DIOT SB ZU
OHWR0083 · revision R5 · the latest of 3 revisions
DIOT SB ZU (OHWR0083 R5) is an open-hardware circuit board from its public repository, designed in Altium Designer: a 14-layer board of 222.4 × 100 mm with 1104 components and 1208 nets, published under CERN-OHL-W-2.0.
review 99% walked · 67 findings: 1 blocker · 43 major · 23 minor (126 items)
Open in the interactive viewerRead the full review

Design review
Probe recorded 67 findings: 1 blocker · 43 major · 23 minor (126 items) on this design. Each is shown here with its reasoning; the full review links every one to the pins, nets and datasheet pages it rests on.
- PowerGroundShortblockerOpen
'GND' carries a power pin (IC30.22) and a ground pin (IC48.3)
Net 'GND' connects a power pin (IC30.22) and a ground pin (IC48.3).
At GND
PS_POR_B is released on P3V3 alone: P0V85, P1V2, P1V8 and VCC_PSPLL have no power-good path, so a rail that fails to start leaves the MPSoC partly powered and out of reset (CERN #385, still in V5)
V5 still releases PS_POR_B on P3V3 alone. The IC30/IC31 power-good outputs for P0V85, P1V2, P1V8 and the MGT rails are unconnected, so if one of those rails fails to start, IC27 still releases the Zynq from reset about 400 ms after P3V3. The PS supplies VCC_PSINTLP/FP, VCCO_PSDDR, VCCO_PSIO0 and VCC_PSPLL are never checked (sheets 13, 25, 26; CERN issue #385).
The power-good chain as wired in V5
- IC30 (IRPS5401) has EN_L/A/B/C/D (pins 28, 42, 44, 51, 53) all on one net, NetIC30_28. Only PG_B (pin 30, P1V8_AUX) is used, as P1V8AUX_PG.
- P1V8AUX_PG drives all five IC31 enables. Of IC31's outputs only PG_A (pin 26, P3V3) is used, as P3V3_PG, which enables IC22 (P2V5) and IC23 (VTT).
- IC27 MAX16025 has IN1 and IN2 (pins 2, 3) tied together on one divider from P3V3 (R176 68k / R178 13k, 0.5 V threshold, so it trips at about 3.12 V). OUT1 (pin 11) is PS_POR_B, with CDLY1 = 100 nF ('400ms' in the sheet 13 note).
``
mermaid flowchart LR N["NetIC30_28<br/>IC30 EN_L/A/B/C/D"] --> A["IC30 rails<br/>P0V85, P1V8_AUX, P1V2, P1V8"] A -- "PG_B only" --> B["P1V8AUX_PG<br/>IC31 EN_L/A/B/C/D"] B --> C["IC31 rails<br/>P3V3, MGT_1V8, MGT_0V9, MGT_1V2, P1V1"] C -- "PG_A only" --> D["P3V3_PG<br/>IC22 P2V5, IC23 VTT"] P["P3V3 via R176/R178"] --> E["IC27 MAX16025 IN1+IN2"] E -- "CDLY1 100 nF" --> F["PS_POR_B released"]``Unmonitored PS supplies
PS supply (DS925 name) Board net Producer Its power-good pin Source VCC_PSINTLP, VCC_PSINTFP, VCC_PSINTFP_DDR P0V85 IC30 loop A + IC29 IC30 PG_A (pin 26): unconnected netlist; UG1075 ball names via bank map VCCO_PSDDR P1V2 IC30 loop C IC30 PG_C (pin 27): unconnected netlist VCCO_PSIO0 (bank 500) P1V8 IC30 loop D IC30 PG_D (pin 16): unconnected netlist; sheet 16 VCC_PSPLL MGT_1V2 through net tie NT7 IC31 loop D IC31 PG_D (pin 16): unconnected netlist; sheet 26 VCCO_PSIO1/2/3 P3V3 IC31 loop A + IC41 PG_A to P3V3_PG; P3V3 also feeds IC27 netlist; sheet 13 Requirement
Item Value Source PS_POR_B "must be asserted Low at power-up and continue to be asserted ... after all the PS supply voltages reach minimum levels" DS925 p.35, Table 37 note 1 FPD "must be powered before PS_POR_B is released" DS925 p.16 IRPS5401 PG pins PG_A 26, PG_B 30, PG_C 27, PG_D 16, PG_L 25 (open drain) IRPS5401 datasheet p.8-9 Shared-fault option "global_fault_en": an MTP bit that, when set, shuts all five outputs on a fault in one IRPS5401 datasheet p.37 IC27 IN1/IN2 monitored inputs, each with its own output MAX16025 datasheet p.6 Why it matters
In normal power-up the IRPS5401 delays (sheet 24) bring every rail up before the 400 ms POR delay runs out. The gap shows when a rail fails to start or trips later. If IC30 loop A, C or D, or IC31 loop B, C or D, stops while P1V8_AUX and P3V3 stay up, the Zynq is partly powered: VCCO and MGTAVTT are live with VCCINT or VCC_PSPLL absent, and PS_POR_B is released or stays high. That breaks DS925 Table 37.
Within each IRPS5401, the outcome depends on global_fault_en in the MTP image. CERN's V5 images (IC30_signed.txt / IC31_signed.txt on the project wiki) are raw register dumps, so this review cannot read that bit. Even with it set, a loop that never reaches its target without tripping a fault still goes unseen, because its PG pin is not used. Nothing in V5 changes this relative to V4 (R4 OBS-29).
Possible fix
Wire the unused open-drain PG outputs (IC30 PG_A/C/D, IC31 PG_B/C/D) into P1V8AUX_PG/P3V3_PG, or into IC27's IN2/EN inputs, so PS_POR_B waits for every PS supply.
USB-JTAG translator IC4 (SN74AVC2T245) drives PS_SRST_B push-pull through R324 0R, against supervisor IC27's open-drain OUT2: with USB connected, the power-on reset delay and the PB3 reset are overridden (sheets 4, 13)
V5 still connects a push-pull translator output to the MPSoC system reset. IC4 B1 drives USB_JTAG.TRSTn, and R324 (0R, fitted) ties that net to PS_SRST_B, which MAX16025 IC27 drives through its open-drain OUT2. With a USB cable connected and no JTAG pod on J7, IC4 is enabled and drives PS_SRST_B high. That overrides IC27's 800 ms power-on hold and the PB3 manual reset, and loads OUT2 beyond what it is specified to sink.
The path
Item Value Source IC4 A1 (pin 8) USB_TRSTn: FT4232H pin 20, R369 10k to P3V3_USB, so it idles high netlist; sheet 4 IC4 DIR1 (pin 10) / OE (pin 2) P3V3_USB (A to B) / EN_USB_JTAG_N netlist; sheet 4 IC4 VCCA / VCCB P3V3_USB / P3V3 netlist EN_USB_JTAG_N R43 10k to P3V3_USB, pulled low by T18. T18 gate EN_USB_JTAG has R336 10k to P3V3; only J7 pin 13 grounds it netlist; sheet 13 IC4 B1 (pin 5) USB_JTAG.TRSTn, R324 0R (fitted, V5 BOM) to PS_SRST_B netlist; sheet 13 PS_SRST_B members IC27 pin 10 (OUT2), IC52 AB27, J7 pin 14, R46 4k7 to P3V3, R45 0R to FMC_JTAG.TRST (J1 d34) netlist IC27 OUT2 timing CDLY2 = C88 220 nF: 'Delay = 800ms' sheet 13 note SN74AVC2T245 drive IOH -12 mA at VCCB 3.0-3.6 V; OE low and DIR high pass A to B SN74AVC2T245 datasheet p.5, p.12 MAX16025 OUT_ open-drain; VOL 0.3 V only specified at 0.5 mA sink; +/-20 mA per pin absolute maximum MAX16025 datasheet p.3, p.2 ``
mermaid flowchart LR F["FT4232H pin 20<br/>USB_TRSTn, R369 pull-up"] --> I["IC4 A1 to B1<br/>push-pull, OE low"] I -- "R324 0R" --> N["PS_SRST_B"] S["IC27 MAX16025 OUT2<br/>open drain, 800 ms delay"] --> N B["PB3 to IC27 EN2"] --> S N --> Z["IC52 AB27 PS_SRST_B"] N -- "R45 0R" --> T["FMC TRST (J1 d34)"]``When it happens
P3V3_USB is present whenever a USB cable is connected (IC3 runs from USB VBUS). If no JTAG pod grounds J7 pin 13, T18 holds IC4's OE low as soon as P3V3 is up. A1 idles high, so B1 drives PS_SRST_B high:
- during the 800 ms that IC27 tries to hold PS_SRST_B low after P3V3 comes up;
- when PB3 (IC27 EN2) is pressed, or on a P3V3 brown-out.
With USB unplugged, P3V3_USB is off and the translator's VCC isolation puts both ports in high impedance (SN74AVC2T245 p.12), so the problem only appears with USB connected.
Why it matters
A push-pull high specified at 12 mA beats an open-drain low specified at 0.5 mA, so the supervisor's system-reset hold-off and manual reset do not work while USB is connected, and the FMC's TRST follows. OUT2 is driven toward its 20 mA pin limit for as long as it tries to assert. This is unchanged from V4 (R4 OBS-35).
Possible fix
Make the USB reset path open-drain: drive an N-FET or an open-drain buffer from USB_TRSTn instead of a push-pull translator output, or fit R324 only when USB-driven reset is wanted.
BZX84C3V3 went end-of-life (Obsolete)
Vendor (DigiKey) lifecycle for BZX84C3V3 (onsemi's listing) changed from "Active" to "Obsolete" since this board was reviewed. Placed as D1, D3. A sourcing/longevity risk; consider a replacement before production.
- concernCON-036majorAgreed
BZX84C3V3 went end-of-life (Obsolete)
[obs#580] BZX84C3V3 went end-of-life (Obsolete) - Vendor (DigiKey) lifecycle for BZX84C3V3 (onsemi's listing) changed from "Active" to "Obsolete" since this board was reviewed. Placed as D1, D3. A sourcing/longevity risk; consider a replacement before production.
- concernCON-019majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_P0] OffBoardCoupling on FMC.LA_P0: ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_P0 for 37.3 mm and puts an estimated 70.5 mV on it (near-end; near-end 70.5 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P0.
- concernCON-018majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_N25] OffBoardCoupling on FMC.LA_N25: ETH_TXD3 runs over each other on L8 and L7 with FMC.LA_N25 for 37.6 mm and puts an estimated 91.6 mV on it (near-end; near-end 91.6 mV), at ETH_TXD3's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N25.
- concernCON-020majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_P25] OffBoardCoupling on FMC.LA_P25: ETH_TX_CTL runs over each other on L8 and L7 with FMC.LA_P25 for 25.1 mm and puts an estimated 42.8 mV on it (near-end; near-end 42.8 mV), at ETH_TX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P25.
- concernCON-021majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_N14] OffBoardCoupling on FMC.LA_N14: ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_N14 for 13.2 mm and puts an estimated 31.1 mV on it (near-end; near-end 31.1 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N14.
- concernCON-026majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT2.LVDS_2_N] OffBoardCoupling on SLOT2.LVDS_2_N: ETH_RX_CTL runs over each other on L8 and L7 with SLOT2.LVDS_2_N for 5 mm and puts an estimated 15.7 mV on it (near-end; near-end 15.7 mV), at ETH_RX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT2.LVDS_2_N has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT2.LVDS_2_N.
- concernCON-029majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT4.LVDS_15_N] OffBoardCoupling on SLOT4.LVDS_15_N: ETH_TXD2 runs over each other on L8 and L7 with SLOT4.LVDS_15_N for 3.9 mm and puts an estimated 15.1 mV on it (near-end; near-end 15.1 mV), at ETH_TXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT4.LVDS_15_N has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT4.LVDS_15_N.
- concernCON-022majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT2.LVDS_2_P] OffBoardCoupling on SLOT2.LVDS_2_P: ETH_RXD2 runs over each other on L8 and L7 with SLOT2.LVDS_2_P for 5.4 mm and puts an estimated 23 mV on it (near-end; near-end 23 mV), at ETH_RXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT2.LVDS_2_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT2.LVDS_2_P.
- concernCON-028majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_N11] OffBoardCoupling on FMC.LA_N11: ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_N11 for 5.1 mm and puts an estimated 15.4 mV on it (near-end; near-end 15.4 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N11 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N11.
- concernCON-030majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_P11] OffBoardCoupling on FMC.LA_P11: ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_P11 for 13.1 mm and puts an estimated 13.2 mV on it (near-end; near-end 13.2 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P11 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P11.
- concernCON-031majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT5.LVDS_15_N] OffBoardCoupling on SLOT5.LVDS_15_N: ETH_RXD0 runs over each other on L8 and L7 with SLOT5.LVDS_15_N for 3.1 mm and puts an estimated 11.9 mV on it (near-end; near-end 11.9 mV), at ETH_RXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT5.LVDS_15_N has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT5.LVDS_15_N.
- concernCON-023majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_P14] OffBoardCoupling on FMC.LA_P14: ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_P14 for 24.5 mm and puts an estimated 21.3 mV on it (near-end; near-end 21.3 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P14.
- concernCON-024majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:FMC.LA_N0] OffBoardCoupling on FMC.LA_N0: ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_N0 for 34.9 mm and puts an estimated 17.8 mV on it (near-end; near-end 17.8 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N0.
- concernCON-025majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT1.LVDS_3_N] OffBoardCoupling on SLOT1.LVDS_3_N: ETH_TX_CTL runs over each other on L8 and L7 with SLOT1.LVDS_3_N for 3.3 mm and puts an estimated 15.7 mV on it (near-end; near-end 15.7 mV), at ETH_TX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_N has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_N.
- concernCON-027majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT1.LVDS_3_P] OffBoardCoupling on SLOT1.LVDS_3_P: ETH_RX_CLK runs over each other on L8 and L7 with SLOT1.LVDS_3_P for 3.2 mm and puts an estimated 15.5 mV on it (near-end; near-end 15.5 mV), at ETH_RX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_P has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_P.
- concernCON-032majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT5.LVDS_15_P] OffBoardCoupling on SLOT5.LVDS_15_P: ETH_RX_CTL runs over each other on L8 and L7 with SLOT5.LVDS_15_P for 2.8 mm and puts an estimated 10.1 mV on it (near-end; near-end 10.1 mV), at ETH_RX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT5.LVDS_15_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT5.LVDS_15_P.
- concernCON-033majorAgreed
OffBoardCoupling
[finding:OffBoardCoupling:SLOT4.LVDS_15_P] OffBoardCoupling on SLOT4.LVDS_15_P: ETH_TXD2 runs over each other on L8 and L7 with SLOT4.LVDS_15_P for 4.4 mm and puts an estimated 7.8 mV on it (near-end; near-end 7.8 mV), at ETH_TXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT4.LVDS_15_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT4.LVDS_15_P.
- concernCON-003majorAgreed
PartLifecycle
[finding:PartLifecycle:T10] PartLifecycle on T10: BSS138L on 19 placements (T10, T11, T12, T13, T14, T15, T16, T17, …): the distributor catalogue reports "Obsolete" for onsemi's BSS138L — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. This board's BOM names ON SEMICONDUCTOR, and that is one line with onsemi: ON Semiconductor completed its acquisition of Fairchild Semiconductor on 2016-09-19 and renamed itself onsemi on 2021-08-05. A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/BSS138L/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2026-01-01, 303 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
- concernCON-002majorAgreed
PartLifecycle
[finding:PartLifecycle:IC6] PartLifecycle on IC6: PCA9548APW on IC6: the distributor catalogue reports "Obsolete" for NXP USA Inc.'s PCA9548APW,112 — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (NXP SEMICONDUCTORS). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/PCA9548APW/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2018-01-24, 3,202 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
- concernCON-004majorAgreed
PartLifecycle
[finding:PartLifecycle:D1] PartLifecycle on D1: BZX84C3V3 on 2 placements (D1, D3): the distributor catalogue reports "Obsolete" for onsemi's BZX84C3V3 — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. This board's BOM names FAIRCHILD SEMICONDUCTOR, and that is one line with onsemi: ON Semiconductor completed its acquisition of Fairchild Semiconductor on 2016-09-19 and renamed itself onsemi on 2021-08-05. A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/BZX84C3V3/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2017-11-19, 3,268 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
- concernCON-005majorAgreed
PartLifecycle
[finding:PartLifecycle:D9] PartLifecycle on D9: BAT54C on D9: the distributor catalogue reports "Obsolete" for onsemi's BAT54C — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. This board's BOM names FAIRCHILD SEMICONDUCTOR, and that is one line with onsemi: ON Semiconductor completed its acquisition of Fairchild Semiconductor on 2016-09-19 and renamed itself onsemi on 2021-08-05. A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/BAT54C/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2018-09-30, 2,953 day(s) ago), so this is a substitution rather than a buy. The distributor reports 54,983 left. No approved alternate is recorded for this position.
- concernCON-008majorAgreed
PartLifecycle
[finding:PartLifecycle:IC18] PartLifecycle on IC18: MT25QU01GBBB1EW9-0SIT on IC18: the distributor catalogue reports "Obsolete" for Micron Technology Inc.'s MT25QU01GBBB1EW9-0SIT — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (MICRON TECHNOLOGY). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/MT25QU01GBBB1EW9-0SIT/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2026-08-31, 61 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
- concernCON-006majorAgreed
PartLifecycle
[finding:PartLifecycle:IC13] PartLifecycle on IC13: MT40A1G16TB-062E:F on 6 placements (IC13, IC14, IC15, IC16, IC17, IC9): the distributor catalogue reports "Obsolete" for Micron Technology Inc.'s MT40A1G16TB-062E:F — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (MICRON TECHNOLOGY). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/MT40A1G16TB-062E:F/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). No approved alternate is recorded for this position.
- concernCON-007majorAgreed
PartLifecycle
[finding:PartLifecycle:IC7] PartLifecycle on IC7: LP5951MF-3.0 on IC7: the distributor catalogue reports "Obsolete" for Texas Instruments's LP5951MF-3.0 — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (TEXAS INSTRUMENTS). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/LP5951MF-3.0/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). No approved alternate is recorded for this position.
- concernCON-034majorAgreed
PS_POR_B is released on P3V3 alone: P0V85, P1V2, P1V8 and VCC_PSPLL have no power-good path, so a rail that fails to start leaves the MPSoC partly powered and out of reset (CERN #385, still in V5)
[obs#560] PS_POR_B is released on P3V3 alone: P0V85, P1V2, P1V8 and VCC_PSPLL have no power-good path, so a rail that fails to start leaves the MPSoC partly powered and out of reset (CERN #385, still in V5) - V5 still releases PS_POR_B on P3V3 alone. The IC30/IC31 power-good outputs for P0V85, P1V2, P1V8 and the MGT rails are unconnected, so if one of those rails fails to start, IC27 still releases the Zynq from reset about 400 ms after P3V3. The PS supplies VCC_PSINTLP/FP, VCCO_PSDDR, VCCO_PSIO0 and VCC_PSPLL are never checked (sheets 13, 25, 26; CERN issue #385).
The power-good chain as wired in V5
- IC30 (IRPS5401) has EN_L/A/B/C/D (pins 28, 42, 44, 51, 53) all on one net, NetIC30_28. Only PG_B (pin 30, P1V8_AUX) is used, as P1V8AUX_PG.
- P1V8AUX_PG drives all five IC31 enables. Of IC31's outputs only PG_A (pin 26, P3V3) is used, as P3V3_PG, which enables IC22 (P2V5) and IC23 (VTT).
- IC27 MAX16025 has IN1 and IN2 (pins 2, 3) tied together on one divider from P3V3 (R176 68k / R178 13k, 0.5 V threshold, so it trips at about 3.12 V). OUT1 (pin 11) is PS_POR_B, with CDLY1 = 100 nF ('400ms' in the sheet 13 note).
``
mermaid flowchart LR N["NetIC30_28<br/>IC30 EN_L/A/B/C/D"] --> A["IC30 rails<br/>P0V85, P1V8_AUX, P1V2, P1V8"] A -- "PG_B only" --> B["P1V8AUX_PG<br/>IC31 EN_L/A/B/C/D"] B --> C["IC31 rails<br/>P3V3, MGT_1V8, MGT_0V9, MGT_1V2, P1V1"] C -- "PG_A only" --> D["P3V3_PG<br/>IC22 P2V5, IC23 VTT"] P["P3V3 via R176/R178"] --> E["IC27 MAX16025 IN1+IN2"] E -- "CDLY1 100 nF" --> F["PS_POR_B released"]``Unmonitored PS supplies
PS supply (DS925 name) Board net Producer Its power-good pin Source VCC_PSINTLP, VCC_PSINTFP, VCC_PSINTFP_DDR P0V85 IC30 loop A + IC29 IC30 PG_A (pin 26): unconnected netlist; UG1075 ball names via bank map VCCO_PSDDR P1V2 IC30 loop C IC30 PG_C (pin 27): unconnected netlist VCCO_PSIO0 (bank 500) P1V8 IC30 loop D IC30 PG_D (pin 16): unconnected netlist; sheet 16 VCC_PSPLL MGT_1V2 through net tie NT7 IC31 loop D IC31 PG_D (pin 16): unconnected netlist; sheet 26 VCCO_PSIO1/2/3 P3V3 IC31 loop A + IC41 PG_A to P3V3_PG; P3V3 also feeds IC27 netlist; sheet 13 Requirement
Item Value Source PS_POR_B "must be asserted Low at power-up and continue to be asserted ... after all the PS supply voltages reach minimum levels" DS925 p.35, Table 37 note 1 FPD "must be powered before PS_POR_B is released" DS925 p.16 IRPS5401 PG pins PG_A 26, PG_B 30, PG_C 27, PG_D 16, PG_L 25 (open drain) IRPS5401 datasheet p.8-9 Shared-fault option "global_fault_en": an MTP bit that, when set, shuts all five outputs on a fault in one IRPS5401 datasheet p.37 IC27 IN1/IN2 monitored inputs, each with its own output MAX16025 datasheet p.6 Why it matters
In normal power-up the IRPS5401 delays (sheet 24) bring every rail up before the 400 ms POR delay runs out. The gap shows when a rail fails to start or trips later. If IC30 loop A, C or D, or IC31 loop B, C or D, stops while P1V8_AUX and P3V3 stay up, the Zynq is partly powered: VCCO and MGTAVTT are live with VCCINT or VCC_PSPLL absent, and PS_POR_B is released or stays high. That breaks DS925 Table 37.
Within each IRPS5401, the outcome depends on global_fault_en in the MTP image. CERN's V5 images (IC30_signed.txt / IC31_signed.txt on the project wiki) are raw register dumps, so this review cannot read that bit. Even with it set, a loop that never reaches its target without tripping a fault still goes unseen, because its PG pin is not used. Nothing in V5 changes this relative to V4 (R4 OBS-29).
Possible fix
Wire the unused open-drain PG outputs (IC30 PG_A/C/D, IC31 PG_B/C/D) into P1V8AUX_PG/P3V3_PG, or into IC27's IN2/EN inputs, so PS_POR_B waits for every PS supply.
- concernCON-012majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L13GND:hole 172.36,131.93|179.78,136.5] ReferencePlaneDiscontinuity on gap:L13GND:hole 172.36,131.93|179.78,136.5: Along 'WR_DAC.SYNC1' on BOTTOM_LAYER, L13GND is a negative layer (copper everywhere inside the profile), and its field is cleared for 2.16 mm under the track at (179.33, 134.14). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. 2 places on 2 net(s) cross it, listed below, the longest first. What cleared it: the hole at x 172.36..179.78, y 131.93..136.5. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'WR_DAC.SYNC1' on BOTTOM_LAYER at (179.33, 134.14), 2.16 mm; other side: none: the signal layer is outermost
- 'WR_DAC.SCLK' on BOTTOM_LAYER at (179.65, 134.12), 2.02 mm; other side: none: the signal layer is outermost
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL13GND%3Ahole%20172.36%2C131.93%7C179.78%2C136.5 serves each with its own layer_image region; all of them lie in x 178.27..180.65, y 132.05..136.18 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-009majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L11PWR:hole 189.45,119.13|189.95,119.63] ReferencePlaneDiscontinuity on gap:L11PWR:hole 189.45,119.13|189.95,119.63: Along 'WR_DAC.DIN' on L12, L11PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.39 mm under the track at (189.79, 119.29). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 189.45..189.95, y 119.13..119.63. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'WR_DAC.DIN' on L12 at (189.79, 119.29), 0.39 mm; other side: no plane on L13GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL11PWR%3Ahole%20189.45%2C119.13%7C189.95%2C119.63 serves each with its own layer_image region; all of them lie in x 188.7..190.98, y 118.1..120.38 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-014majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L4PWR:hole 225.01,107.24|225.51,107.74] ReferencePlaneDiscontinuity on gap:L4PWR:hole 225.01,107.24|225.51,107.74: Along 'SLOT1.LVDS_2_P' on L5, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.5 mm under the track at (225.26, 107.36). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 225.01..225.51, y 107.24..107.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT1.LVDS_2_P' on L5 at (225.26, 107.36), 0.5 mm; other side: no plane on L6GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20225.01%2C107.24%7C225.51%2C107.74 serves each with its own layer_image region; all of them lie in x 224.26..226.45, y 106.15..108.49 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-010majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L11PWR:hole 189.68,121.38|190.18,121.88] ReferencePlaneDiscontinuity on gap:L11PWR:hole 189.68,121.38|190.18,121.88: Along 'WR_DAC.SYNC1' on L12, L11PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.41 mm under the track at (189.93, 121.75). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 189.68..190.18, y 121.38..121.88. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'WR_DAC.SYNC1' on L12 at (189.93, 121.75), 0.41 mm; other side: no plane on L13GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL11PWR%3Ahole%20189.68%2C121.38%7C190.18%2C121.88 serves each with its own layer_image region; all of them lie in x 188.93..190.97, y 120.63..123.02 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-011majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L11PWR:hole 233.01,110.24|233.51,110.74] ReferencePlaneDiscontinuity on gap:L11PWR:hole 233.01,110.24|233.51,110.74: Along 'SLOT4.LVDS_0_P' on L12, L11PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.56 mm under the track at (233.35, 110.4). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 233.01..233.51, y 110.24..110.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT4.LVDS_0_P' on L12 at (233.35, 110.4), 0.56 mm; other side: no plane on L13GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL11PWR%3Ahole%20233.01%2C110.24%7C233.51%2C110.74 serves each with its own layer_image region; all of them lie in x 232.26..234.68, y 109.15..111.49 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-013majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L4PWR:hole 225.01,106.24|225.51,106.74] ReferencePlaneDiscontinuity on gap:L4PWR:hole 225.01,106.24|225.51,106.74: Along 'SLOT1.LVDS_2_N' on L5, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.54 mm under the track at (225.26, 106.61). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 225.01..225.51, y 106.24..106.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT1.LVDS_2_N' on L5 at (225.26, 106.61), 0.54 mm; other side: no plane on L6GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20225.01%2C106.24%7C225.51%2C106.74 serves each with its own layer_image region; all of them lie in x 224.26..226.45, y 105.49..107.83 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-015majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L4PWR:hole 230.01,139.24|230.51,139.74] ReferencePlaneDiscontinuity on gap:L4PWR:hole 230.01,139.24|230.51,139.74: Along 'SLOT8.LVDS_10_P' on L5, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.33 mm under the track at (230.38, 139.49). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 230.01..230.51, y 139.24..139.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT8.LVDS_10_P' on L5 at (230.38, 139.49), 0.33 mm; other side: no plane on L6GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20230.01%2C139.24%7C230.51%2C139.74 serves each with its own layer_image region; all of them lie in x 229.26..231.58, y 138.46..140.49 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-016majorAgreed
ReferencePlaneDiscontinuity
[finding:ReferencePlaneDiscontinuity:gap:L9GND:hole 85.5,116.35|124.1,118.35] ReferencePlaneDiscontinuity on gap:L9GND:hole 85.5,116.35|124.1,118.35: Along 'ETH_TX_CLK' on L8, L9GND is a negative layer (copper everywhere inside the profile), and its field is cleared for 8.12 mm under the track at (119.44, 116.68). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. 2 places on 2 net(s) cross it, listed below, the longest first. What cleared it: the hole at x 85.5..124.1, y 116.35..118.35. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'ETH_TX_CLK' on L8 at (119.44, 116.68), 8.12 mm; other side: no plane on L7
- 'ETH_TXD1' on L8 at (120, 116.42), 6.45 mm; other side: no plane on L7
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL9GND%3Ahole%2085.5%2C116.35%7C124.1%2C118.35 serves each with its own layer_image region; all of them lie in x 115.2..125.18, y 115.35..117.68 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
- concernCON-017majorAgreed
SignalNearPlaneEdge
[finding:SignalNearPlaneEdge:ETH_TX_CLK] SignalNearPlaneEdge on ETH_TX_CLK: 'ETH_TX_CLK' runs within 0.31 mm (1.6H) of the edge of the P3V3 plane on L10 at 1 place(s). Its edge is 1000 ps 10-90 % (group: 1 ns 10-90 %, signal group 'Ethernet' () by : STARTING VALUE, not a sourced figure: Probe's starting edge for Ethernet (1 ns 10-90 %) and 3.3 V, created when the name suggestions were accepted; replace it with the drivers' datasheet edge), knee 500 MHz. Where the plane ends beside a track, the return current that would have spread past the edge is crowded into it, and the edge's fringing field radiates. Sites: on L8 over P3V3 on L10 at (116.53, 116.68): the edge 0.31 mm from the centreline at (116.53, 116.99), for 7.39 mm along the track, H = 0.2 mm, 18 % of the return beyond the edge; other side: no plane on that side. 3H from the plane's edge is a rule of thumb, declared as one: on an infinite plane the share of the return beyond an edge D from the centreline is 1/2 - atan(D/H)/pi (the microstrip current distribution 1/(1+(x/H)^2)), a tenth at 3H. Heights are from the export's own stackup rows. Moving the track inward, or extending the plane past it, closes it.
- concernCON-035majorAgreed
USB-JTAG translator IC4 (SN74AVC2T245) drives PS_SRST_B push-pull through R324 0R, against supervisor IC27's open-drain OUT2: with USB connected, the power-on reset delay and the PB3 reset are overridden (sheets 4, 13)
[obs#566] USB-JTAG translator IC4 (SN74AVC2T245) drives PS_SRST_B push-pull through R324 0R, against supervisor IC27's open-drain OUT2: with USB connected, the power-on reset delay and the PB3 reset are overridden (sheets 4, 13) - V5 still connects a push-pull translator output to the MPSoC system reset. IC4 B1 drives USB_JTAG.TRSTn, and R324 (0R, fitted) ties that net to PS_SRST_B, which MAX16025 IC27 drives through its open-drain OUT2. With a USB cable connected and no JTAG pod on J7, IC4 is enabled and drives PS_SRST_B high. That overrides IC27's 800 ms power-on hold and the PB3 manual reset, and loads OUT2 beyond what it is specified to sink.
The path
Item Value Source IC4 A1 (pin 8) USB_TRSTn: FT4232H pin 20, R369 10k to P3V3_USB, so it idles high netlist; sheet 4 IC4 DIR1 (pin 10) / OE (pin 2) P3V3_USB (A to B) / EN_USB_JTAG_N netlist; sheet 4 IC4 VCCA / VCCB P3V3_USB / P3V3 netlist EN_USB_JTAG_N R43 10k to P3V3_USB, pulled low by T18. T18 gate EN_USB_JTAG has R336 10k to P3V3; only J7 pin 13 grounds it netlist; sheet 13 IC4 B1 (pin 5) USB_JTAG.TRSTn, R324 0R (fitted, V5 BOM) to PS_SRST_B netlist; sheet 13 PS_SRST_B members IC27 pin 10 (OUT2), IC52 AB27, J7 pin 14, R46 4k7 to P3V3, R45 0R to FMC_JTAG.TRST (J1 d34) netlist IC27 OUT2 timing CDLY2 = C88 220 nF: 'Delay = 800ms' sheet 13 note SN74AVC2T245 drive IOH -12 mA at VCCB 3.0-3.6 V; OE low and DIR high pass A to B SN74AVC2T245 datasheet p.5, p.12 MAX16025 OUT_ open-drain; VOL 0.3 V only specified at 0.5 mA sink; +/-20 mA per pin absolute maximum MAX16025 datasheet p.3, p.2 ``
mermaid flowchart LR F["FT4232H pin 20<br/>USB_TRSTn, R369 pull-up"] --> I["IC4 A1 to B1<br/>push-pull, OE low"] I -- "R324 0R" --> N["PS_SRST_B"] S["IC27 MAX16025 OUT2<br/>open drain, 800 ms delay"] --> N B["PB3 to IC27 EN2"] --> S N --> Z["IC52 AB27 PS_SRST_B"] N -- "R45 0R" --> T["FMC TRST (J1 d34)"]``When it happens
P3V3_USB is present whenever a USB cable is connected (IC3 runs from USB VBUS). If no JTAG pod grounds J7 pin 13, T18 holds IC4's OE low as soon as P3V3 is up. A1 idles high, so B1 drives PS_SRST_B high:
- during the 800 ms that IC27 tries to hold PS_SRST_B low after P3V3 comes up;
- when PB3 (IC27 EN2) is pressed, or on a P3V3 brown-out.
With USB unplugged, P3V3_USB is off and the translator's VCC isolation puts both ports in high impedance (SN74AVC2T245 p.12), so the problem only appears with USB connected.
Why it matters
A push-pull high specified at 12 mA beats an open-drain low specified at 0.5 mA, so the supervisor's system-reset hold-off and manual reset do not work while USB is connected, and the FMC's TRST follows. OUT2 is driven toward its 20 mA pin limit for as long as it tries to assert. This is unchanged from V4 (R4 OBS-35).
Possible fix
Make the USB reset path open-drain: drive an N-FET or an open-drain buffer from USB_TRSTn instead of a push-pull translator output, or fit R324 only when USB-driven reset is wanted.
- concernCON-001majorAgreed
[eol:BZX84C3V3:Obsolete] BZX84C3V3 lifecycle changed from "Active" to "Obsolete" since this board was reviewed: OHWR0083 R5 DIOT SB ZU. A sourcing and longevity risk. Should this be replaced before the next build?
- SignalNearPlaneEdgemajorAgreed
'ETH_TX_CLK' runs within 0.31 mm (1.6H) of the edge of the P3V3 plane on L10 at 1 place(s)
'ETH_TX_CLK' runs within 0.31 mm (1.6H) of the edge of the P3V3 plane on L10 at 1 place(s). Its edge is 1000 ps 10-90 % (group: 1 ns 10-90 %, signal group 'Ethernet' () by : STARTING VALUE, not a sourced figure: Probe's starting edge for Ethernet (1 ns 10-90 %) and 3.3 V, created when the name suggestions were accepted; replace it with the drivers' datasheet edge), knee 500 MHz. Where the plane ends beside a track, the return current that would have spread past the edge is crowded into it, and the edge's fringing field radiates. Sites: on L8 over P3V3 on L10 at (116.53, 116.68): the edge 0.31 mm from the centreline at (116.53, 116.99), for 7.39 mm along the track, H = 0.2 mm, 18 % of the return beyond the edge; other side: no plane on that side. 3H from the plane's edge is a rule of thumb, declared as one: on an infinite plane the share of the return beyond an edge D from the centreline is 1/2 - atan(D/H)/pi (the microstrip current distribution 1/(1+(x/H)^2)), a tenth at 3H. Heights are from the export's own stackup rows. Moving the track inward, or extending the plane past it, closes it.
At ETH_TX_CLK
- OffBoardCouplingmajorAgreed
OffBoardCoupling: 42 nets×42
Agreed · evaluation@esharp.se — Real layout weakness as read from the copper; tracked with its siblings in OBS-637 (confirmed, minor). Magnitude not quantified: the edge rates are Probe's assumed values, not sourced. Owner decision approved by Daniel Rhodin, E-Sharp AB, 2026-11-02 (drafted by Claude from the walker's evidence).
At FMC.LA_N0, FMC.LA_N11, FMC.LA_N14, FMC.LA_N25, FMC.LA_P0, FMC.LA_P11 and 10 more
Each finding, with its own headline
FMC.LA_N0FMC.LA_N0 leaves the board at J1 carrying 17.8 mV from ETH_TXD1, coupled over 34.9 mm
ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_N0 for 34.9 mm and puts an estimated 17.8 mV on it (near-end; near-end 17.8 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N0.
FMC.LA_N0FMC.LA_N0 leaves the board at J1 carrying 16 mV from ETH_RXD2, coupled over 5.7 mm
ETH_RXD2 runs over each other on L8 and L7 with FMC.LA_N0 for 5.7 mm and puts an estimated 16 mV on it (near-end; near-end 16 mV), at ETH_RXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N0.
FMC.LA_N0FMC.LA_N0 leaves the board at J1 carrying 11.7 mV from ETH_TX_CLK, coupled over 33.7 mm
ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_N0 for 33.7 mm and puts an estimated 11.7 mV on it (near-end; near-end 11.7 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N0.
FMC.LA_N0FMC.LA_N0 leaves the board at J1 carrying 4.8 mV from ETH_TXD0, coupled over 29.2 mm
ETH_TXD0 runs over each other on L8 and L7 with FMC.LA_N0 for 29.2 mm and puts an estimated 4.8 mV on it (near-end; near-end 4.8 mV), at ETH_TXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N0.
FMC.LA_N11FMC.LA_N11 leaves the board at J1 carrying 15.4 mV from ETH_TX_CLK, coupled over 5.1 mm
ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_N11 for 5.1 mm and puts an estimated 15.4 mV on it (near-end; near-end 15.4 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N11 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N11.
FMC.LA_N14FMC.LA_N14 leaves the board at J1 carrying 31.1 mV from ETH_TXD1, coupled over 13.2 mm
ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_N14 for 13.2 mm and puts an estimated 31.1 mV on it (near-end; near-end 31.1 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N14.
FMC.LA_N14FMC.LA_N14 leaves the board at J1 carrying 14.6 mV from ETH_TX_CLK, coupled over 23.8 mm
ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_N14 for 23.8 mm and puts an estimated 14.6 mV on it (near-end; near-end 14.6 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N14.
FMC.LA_N14FMC.LA_N14 leaves the board at J1 carrying 12.6 mV from ETH_TX_CTL, coupled over 7.6 mm
ETH_TX_CTL runs over each other on L8 and L7 with FMC.LA_N14 for 7.6 mm and puts an estimated 12.6 mV on it (near-end; near-end 12.6 mV), at ETH_TX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N14.
FMC.LA_N25FMC.LA_N25 leaves the board at J1 carrying 91.6 mV from ETH_TXD3, coupled over 37.6 mm
ETH_TXD3 runs over each other on L8 and L7 with FMC.LA_N25 for 37.6 mm and puts an estimated 91.6 mV on it (near-end; near-end 91.6 mV), at ETH_TXD3's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N25.
FMC.LA_N25FMC.LA_N25 leaves the board at J1 carrying 47 mV from ETH_TXD0, coupled over 36.1 mm
ETH_TXD0 runs over each other on L8 and L7 with FMC.LA_N25 for 36.1 mm and puts an estimated 47 mV on it (near-end; near-end 47 mV), at ETH_TXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N25.
FMC.LA_N25FMC.LA_N25 leaves the board at J1 carrying 8.8 mV from ETH_TX_CTL, coupled over 24.7 mm
ETH_TX_CTL runs over each other on L8 and L7 with FMC.LA_N25 for 24.7 mm and puts an estimated 8.8 mV on it (near-end; near-end 8.8 mV), at ETH_TX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N25.
FMC.LA_N25FMC.LA_N25 leaves the board at J1 carrying 5.9 mV from ETH_RX_CTL, coupled over 3.7 mm
ETH_RX_CTL runs over each other on L8 and L7 with FMC.LA_N25 for 3.7 mm and puts an estimated 5.9 mV on it (near-end; near-end 5.9 mV), at ETH_RX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_N25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_N25.
FMC.LA_P0FMC.LA_P0 leaves the board at J1 carrying 70.5 mV from ETH_TXD1, coupled over 37.3 mm
ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_P0 for 37.3 mm and puts an estimated 70.5 mV on it (near-end; near-end 70.5 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P0.
FMC.LA_P0FMC.LA_P0 leaves the board at J1 carrying 18.8 mV from ETH_TXD0, coupled over 29.2 mm
ETH_TXD0 runs over each other on L8 and L7 with FMC.LA_P0 for 29.2 mm and puts an estimated 18.8 mV on it (near-end; near-end 18.8 mV), at ETH_TXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P0.
FMC.LA_P0FMC.LA_P0 leaves the board at J1 carrying 6.9 mV from ETH_TX_CLK, coupled over 33.7 mm
ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_P0 for 33.7 mm and puts an estimated 6.9 mV on it (near-end; near-end 6.9 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P0.
FMC.LA_P0FMC.LA_P0 leaves the board at J1 carrying 3.6 mV from ETH_RXD2, coupled over 5.8 mm
ETH_RXD2 runs over each other on L8 and L7 with FMC.LA_P0 for 5.8 mm and puts an estimated 3.6 mV on it (near-end; near-end 3.6 mV), at ETH_RXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P0 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P0.
FMC.LA_P11FMC.LA_P11 leaves the board at J1 carrying 13.2 mV from ETH_TX_CLK, coupled over 13.1 mm
ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_P11 for 13.1 mm and puts an estimated 13.2 mV on it (near-end; near-end 13.2 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P11 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P11.
FMC.LA_P11FMC.LA_P11 leaves the board at J1 carrying 5.5 mV from ETH_TXD1, coupled over 6.1 mm
ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_P11 for 6.1 mm and puts an estimated 5.5 mV on it (near-end; near-end 5.5 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P11 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P11.
FMC.LA_P14FMC.LA_P14 leaves the board at J1 carrying 21.3 mV from ETH_TX_CLK, coupled over 24.5 mm
ETH_TX_CLK runs over each other on L8 and L7 with FMC.LA_P14 for 24.5 mm and puts an estimated 21.3 mV on it (near-end; near-end 21.3 mV), at ETH_TX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P14.
FMC.LA_P14FMC.LA_P14 leaves the board at J1 carrying 13.3 mV from ETH_TXD2, coupled over 6.9 mm
ETH_TXD2 runs over each other on L8 and L7 with FMC.LA_P14 for 6.9 mm and puts an estimated 13.3 mV on it (near-end; near-end 13.3 mV), at ETH_TXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P14.
FMC.LA_P14FMC.LA_P14 leaves the board at J1 carrying 10.8 mV from ETH_TXD1, coupled over 17.1 mm
ETH_TXD1 runs over each other on L8 and L7 with FMC.LA_P14 for 17.1 mm and puts an estimated 10.8 mV on it (near-end; near-end 10.8 mV), at ETH_TXD1's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P14 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P14.
FMC.LA_P25FMC.LA_P25 leaves the board at J1 carrying 42.8 mV from ETH_TX_CTL, coupled over 25.1 mm
ETH_TX_CTL runs over each other on L8 and L7 with FMC.LA_P25 for 25.1 mm and puts an estimated 42.8 mV on it (near-end; near-end 42.8 mV), at ETH_TX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P25.
FMC.LA_P25FMC.LA_P25 leaves the board at J1 carrying 38.4 mV from ETH_TXD0, coupled over 36.2 mm
ETH_TXD0 runs over each other on L8 and L7 with FMC.LA_P25 for 36.2 mm and puts an estimated 38.4 mV on it (near-end; near-end 38.4 mV), at ETH_TXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P25.
FMC.LA_P25FMC.LA_P25 leaves the board at J1 carrying 25.8 mV from ETH_TXD3, coupled over 38.5 mm
ETH_TXD3 runs over each other on L8 and L7 with FMC.LA_P25 for 38.5 mm and puts an estimated 25.8 mV on it (near-end; near-end 25.8 mV), at ETH_TXD3's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). FMC.LA_P25 has a contact on J1, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=FMC.LA_P25.
SLOT1.LVDS_3_NSLOT1.LVDS_3_N leaves the board at J14 carrying 15.7 mV from ETH_TX_CTL, coupled over 3.3 mm
ETH_TX_CTL runs over each other on L8 and L7 with SLOT1.LVDS_3_N for 3.3 mm and puts an estimated 15.7 mV on it (near-end; near-end 15.7 mV), at ETH_TX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_N has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_N.
SLOT1.LVDS_3_NSLOT1.LVDS_3_N leaves the board at J14 carrying 14.5 mV from ETH_RXD0, coupled over 3.2 mm
ETH_RXD0 runs over each other on L8 and L7 with SLOT1.LVDS_3_N for 3.2 mm and puts an estimated 14.5 mV on it (near-end; near-end 14.5 mV), at ETH_RXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_N has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_N.
SLOT1.LVDS_3_NSLOT1.LVDS_3_N leaves the board at J14 carrying 6.5 mV from ETH_RX_CLK, coupled over 3.2 mm
ETH_RX_CLK runs over each other on L8 and L7 with SLOT1.LVDS_3_N for 3.2 mm and puts an estimated 6.5 mV on it (near-end; near-end 6.5 mV), at ETH_RX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_N has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_N.
SLOT1.LVDS_3_NSLOT1.LVDS_3_N leaves the board at J14 carrying 5 mV from ETH_TXD3, coupled over 4.5 mm
ETH_TXD3 runs over each other on L8 and L7 with SLOT1.LVDS_3_N for 4.5 mm and puts an estimated 5 mV on it (near-end; near-end 5 mV), at ETH_TXD3's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_N has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_N.
SLOT1.LVDS_3_NSLOT1.LVDS_3_N leaves the board at J14 carrying 4.1 mV from ETH_TXD0, coupled over 3.2 mm
ETH_TXD0 runs over each other on L8 and L7 with SLOT1.LVDS_3_N for 3.2 mm and puts an estimated 4.1 mV on it (near-end; near-end 4.1 mV), at ETH_TXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_N has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_N.
SLOT1.LVDS_3_PSLOT1.LVDS_3_P leaves the board at J14 carrying 15.5 mV from ETH_RX_CLK, coupled over 3.2 mm
ETH_RX_CLK runs over each other on L8 and L7 with SLOT1.LVDS_3_P for 3.2 mm and puts an estimated 15.5 mV on it (near-end; near-end 15.5 mV), at ETH_RX_CLK's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_P has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_P.
SLOT1.LVDS_3_PSLOT1.LVDS_3_P leaves the board at J14 carrying 13.8 mV from ETH_TXD3, coupled over 4.5 mm
ETH_TXD3 runs over each other on L8 and L7 with SLOT1.LVDS_3_P for 4.5 mm and puts an estimated 13.8 mV on it (near-end; near-end 13.8 mV), at ETH_TXD3's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_P has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_P.
SLOT1.LVDS_3_PSLOT1.LVDS_3_P leaves the board at J14 carrying 6.8 mV from ETH_TX_CTL, coupled over 3.3 mm
ETH_TX_CTL runs over each other on L8 and L7 with SLOT1.LVDS_3_P for 3.3 mm and puts an estimated 6.8 mV on it (near-end; near-end 6.8 mV), at ETH_TX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_P has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_P.
SLOT1.LVDS_3_PSLOT1.LVDS_3_P leaves the board at J14 carrying 5 mV from ETH_RXD0, coupled over 3.2 mm
ETH_RXD0 runs over each other on L8 and L7 with SLOT1.LVDS_3_P for 3.2 mm and puts an estimated 5 mV on it (near-end; near-end 5 mV), at ETH_RXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT1.LVDS_3_P has a contact on J14, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT1.LVDS_3_P.
SLOT2.LVDS_2_NSLOT2.LVDS_2_N leaves the board at J33 carrying 15.7 mV from ETH_RX_CTL, coupled over 5 mm
ETH_RX_CTL runs over each other on L8 and L7 with SLOT2.LVDS_2_N for 5 mm and puts an estimated 15.7 mV on it (near-end; near-end 15.7 mV), at ETH_RX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT2.LVDS_2_N has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT2.LVDS_2_N.
SLOT2.LVDS_2_NSLOT2.LVDS_2_N leaves the board at J33 carrying 9.3 mV from ETH_RXD2, coupled over 5.1 mm
ETH_RXD2 runs over each other on L8 and L7 with SLOT2.LVDS_2_N for 5.1 mm and puts an estimated 9.3 mV on it (near-end; near-end 9.3 mV), at ETH_RXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT2.LVDS_2_N has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT2.LVDS_2_N.
SLOT2.LVDS_2_PSLOT2.LVDS_2_P leaves the board at J33 carrying 23 mV from ETH_RXD2, coupled over 5.4 mm
ETH_RXD2 runs over each other on L8 and L7 with SLOT2.LVDS_2_P for 5.4 mm and puts an estimated 23 mV on it (near-end; near-end 23 mV), at ETH_RXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT2.LVDS_2_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT2.LVDS_2_P.
SLOT2.LVDS_2_PSLOT2.LVDS_2_P leaves the board at J33 carrying 4.2 mV from ETH_RX_CTL, coupled over 5.3 mm
ETH_RX_CTL runs over each other on L8 and L7 with SLOT2.LVDS_2_P for 5.3 mm and puts an estimated 4.2 mV on it (near-end; near-end 4.2 mV), at ETH_RX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT2.LVDS_2_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT2.LVDS_2_P.
SLOT4.LVDS_15_NSLOT4.LVDS_15_N leaves the board at J33 carrying 15.1 mV from ETH_TXD2, coupled over 3.9 mm
ETH_TXD2 runs over each other on L8 and L7 with SLOT4.LVDS_15_N for 3.9 mm and puts an estimated 15.1 mV on it (near-end; near-end 15.1 mV), at ETH_TXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT4.LVDS_15_N has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT4.LVDS_15_N.
SLOT4.LVDS_15_PSLOT4.LVDS_15_P leaves the board at J33 carrying 7.8 mV from ETH_TXD2, coupled over 4.4 mm
ETH_TXD2 runs over each other on L8 and L7 with SLOT4.LVDS_15_P for 4.4 mm and puts an estimated 7.8 mV on it (near-end; near-end 7.8 mV), at ETH_TXD2's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT4.LVDS_15_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT4.LVDS_15_P.
SLOT5.LVDS_15_NSLOT5.LVDS_15_N leaves the board at J33 carrying 11.9 mV from ETH_RXD0, coupled over 3.1 mm
ETH_RXD0 runs over each other on L8 and L7 with SLOT5.LVDS_15_N for 3.1 mm and puts an estimated 11.9 mV on it (near-end; near-end 11.9 mV), at ETH_RXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT5.LVDS_15_N has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT5.LVDS_15_N.
SLOT5.LVDS_15_PSLOT5.LVDS_15_P leaves the board at J33 carrying 10.1 mV from ETH_RX_CTL, coupled over 2.8 mm
ETH_RX_CTL runs over each other on L8 and L7 with SLOT5.LVDS_15_P for 2.8 mm and puts an estimated 10.1 mV on it (near-end; near-end 10.1 mV), at ETH_RX_CTL's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT5.LVDS_15_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT5.LVDS_15_P.
SLOT5.LVDS_15_PSLOT5.LVDS_15_P leaves the board at J33 carrying 3 mV from ETH_RXD0, coupled over 3 mm
ETH_RXD0 runs over each other on L8 and L7 with SLOT5.LVDS_15_P for 3 mm and puts an estimated 3 mV on it (near-end; near-end 3 mV), at ETH_RXD0's edge 1 ns 10-90 % (group) and a 3.3 V swing (group). SLOT5.LVDS_15_P has a contact on J33, so the noise leaves the board with it onto whatever is plugged in there. Moving the two apart, putting a plane between them, or routing the internal net away from the connector's nets is the change. The threshold is the screen's 3 mV listing floor, not a new limit. The run and its sections are in GET crosstalk?net=SLOT5.LVDS_15_P.
- PartLifecyclemajorAgreed
PartLifecycle: 7 part numbers×7
7 Obsolete · on 31 placements
At D1, D9, IC13, IC18, IC6, IC7, T10
Each finding, with its own headline
D1BZX84C3V3 on 2 placements (D1, D3): the distributor catalogue reports "Obsolete" for onsemi's BZX84C3V3
BZX84C3V3 on 2 placements (D1, D3): the distributor catalogue reports "Obsolete" for onsemi's BZX84C3V3 — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. This board's BOM names FAIRCHILD SEMICONDUCTOR, and that is one line with onsemi: ON Semiconductor completed its acquisition of Fairchild Semiconductor on 2016-09-19 and renamed itself onsemi on 2021-08-05. A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/BZX84C3V3/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2017-11-19, 3,283 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
D9BAT54C on D9: the distributor catalogue reports "Obsolete" for onsemi's BAT54C
BAT54C on D9: the distributor catalogue reports "Obsolete" for onsemi's BAT54C — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. This board's BOM names FAIRCHILD SEMICONDUCTOR, and that is one line with onsemi: ON Semiconductor completed its acquisition of Fairchild Semiconductor on 2016-09-19 and renamed itself onsemi on 2021-08-05. A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/BAT54C/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2018-09-30, 2,968 day(s) ago), so this is a substitution rather than a buy. The distributor reports 54,983 left. No approved alternate is recorded for this position.
IC13MT40A1G16TB-062E:F on 6 placements (IC13, IC14, IC15, IC16, IC17, IC9): the distributor catalogue reports "Obsolete" for Micron Technology Inc.'s MT40A1G16TB-062E:F
MT40A1G16TB-062E:F on 6 placements (IC13, IC14, IC15, IC16, IC17, IC9): the distributor catalogue reports "Obsolete" for Micron Technology Inc.'s MT40A1G16TB-062E:F — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (MICRON TECHNOLOGY). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/MT40A1G16TB-062E:F/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). No approved alternate is recorded for this position.
IC18MT25QU01GBBB1EW9-0SIT on IC18: the distributor catalogue reports "Obsolete" for Micron Technology Inc.'s MT25QU01GBBB1EW9-0SIT
MT25QU01GBBB1EW9-0SIT on IC18: the distributor catalogue reports "Obsolete" for Micron Technology Inc.'s MT25QU01GBBB1EW9-0SIT — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (MICRON TECHNOLOGY). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/MT25QU01GBBB1EW9-0SIT/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2026-08-31, 76 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
IC6PCA9548APW on IC6: the distributor catalogue reports "Obsolete" for NXP USA Inc.'s PCA9548APW,112
PCA9548APW on IC6: the distributor catalogue reports "Obsolete" for NXP USA Inc.'s PCA9548APW,112 — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (NXP SEMICONDUCTORS). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/PCA9548APW/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2018-01-24, 3,217 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
IC7LP5951MF-3.0 on IC7: the distributor catalogue reports "Obsolete" for Texas Instruments's LP5951MF-3.0
LP5951MF-3.0 on IC7: the distributor catalogue reports "Obsolete" for Texas Instruments's LP5951MF-3.0 — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. That is the maker this board's BOM names (TEXAS INSTRUMENTS). A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/LP5951MF-3.0/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). No approved alternate is recorded for this position.
T10BSS138L on 19 placements (T10, T11, T12, T13, T14, T15, T16, T17, …): the distributor catalogue reports "Obsolete" for onsemi's BSS138L
BSS138L on 19 placements (T10, T11, T12, T13, T14, T15, T16, T17, …): the distributor catalogue reports "Obsolete" for onsemi's BSS138L — a sourcing/longevity risk to confirm against the MANUFACTURER before acting. This board's BOM names ON SEMICONDUCTOR, and that is one line with onsemi: ON Semiconductor completed its acquisition of Fairchild Semiconductor on 2016-09-19 and renamed itself onsemi on 2021-08-05. A distributor delisting is not the same claim as the part being unavailable; check the vendor's own lifecycle/product-change notice for the status and any named successor. Re-read the source: GET /api/odb/parts/BSS138L/providers (which provider said it, the maker and part number it matched, its own words, and when we asked). THE LAST BUY DATE HAS PASSED (2026-01-01, 318 day(s) ago), so this is a substitution rather than a buy. The distributor reports NONE left. No approved alternate is recorded for this position.
- ReferencePlaneDiscontinuitymajorAgreed
ReferencePlaneDiscontinuity: 8 boards×8
Agreed · evaluation@esharp.se — Real layout weakness as read from the copper; tracked with its siblings in OBS-637 (confirmed, minor). Magnitude not quantified: the edge rates are Probe's assumed values, not sourced. Owner decision approved by Daniel Rhodin, E-Sharp AB, 2026-11-02 (drafted by Claude from the walker's evidence).
At gap:L11PWR:hole 189.45,119.13|189.95,119.63, gap:L11PWR:hole 189.68,121.38|190.18,121.88, gap:L11PWR:hole 233.01,110.24|233.51,110.74, gap:L13GND:hole 172.36,131.93|179.78,136.5, gap:L4PWR:hole 225.01,106.24|225.51,106.74, gap:L4PWR:hole 225.01,107.24|225.51,107.74, gap:L4PWR:hole 230.01,139.24|230.51,139.74, gap:L9GND:hole 85.5,116.35|124.1,118.35
Each finding, with its own headline
gap:L11PWR:hole 189.45,119.13|189.95,119.63'WR_DAC.DIN' crosses one gap in L11PWR's negative-plane field at 1 place, the longest 0.39 mm (fastest: WR_DAC.DIN at 100 ps 10-90 %, ASSUMED)
Along 'WR_DAC.DIN' on L12, L11PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.39 mm under the track at (189.79, 119.29). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 189.45..189.95, y 119.13..119.63. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'WR_DAC.DIN' on L12 at (189.79, 119.29), 0.39 mm; other side: no plane on L13GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL11PWR%3Ahole%20189.45%2C119.13%7C189.95%2C119.63 serves each with its own layer_image region; all of them lie in x 188.7..190.98, y 118.1..120.38 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L11PWR:hole 189.68,121.38|190.18,121.88'WR_DAC.SYNC1' crosses one gap in L11PWR's negative-plane field at 1 place, the longest 0.41 mm (fastest: WR_DAC.SYNC1 at 100 ps 10-90 %, ASSUMED)
Along 'WR_DAC.SYNC1' on L12, L11PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.41 mm under the track at (189.93, 121.75). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 189.68..190.18, y 121.38..121.88. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'WR_DAC.SYNC1' on L12 at (189.93, 121.75), 0.41 mm; other side: no plane on L13GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL11PWR%3Ahole%20189.68%2C121.38%7C190.18%2C121.88 serves each with its own layer_image region; all of them lie in x 188.93..190.97, y 120.63..123.02 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L11PWR:hole 233.01,110.24|233.51,110.74'SLOT4.LVDS_0_P' crosses one gap in L11PWR's negative-plane field at 1 place, the longest 0.56 mm (fastest: SLOT4.LVDS_0_P at 100 ps 10-90 %, ASSUMED)
Along 'SLOT4.LVDS_0_P' on L12, L11PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.56 mm under the track at (233.35, 110.4). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 233.01..233.51, y 110.24..110.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT4.LVDS_0_P' on L12 at (233.35, 110.4), 0.56 mm; other side: no plane on L13GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL11PWR%3Ahole%20233.01%2C110.24%7C233.51%2C110.74 serves each with its own layer_image region; all of them lie in x 232.26..234.68, y 109.15..111.49 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L13GND:hole 172.36,131.93|179.78,136.52 nets cross one gap in L13GND's negative-plane field at 2 places, the longest 2.16 mm (fastest: WR_DAC.SYNC1 at 100 ps 10-90 %, ASSUMED)
Along 'WR_DAC.SYNC1' on BOTTOM_LAYER, L13GND is a negative layer (copper everywhere inside the profile), and its field is cleared for 2.16 mm under the track at (179.33, 134.14). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. 2 places on 2 net(s) cross it, listed below, the longest first. What cleared it: the hole at x 172.36..179.78, y 131.93..136.5. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'WR_DAC.SYNC1' on BOTTOM_LAYER at (179.33, 134.14), 2.16 mm; other side: none: the signal layer is outermost
- 'WR_DAC.SCLK' on BOTTOM_LAYER at (179.65, 134.12), 2.02 mm; other side: none: the signal layer is outermost
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL13GND%3Ahole%20172.36%2C131.93%7C179.78%2C136.5 serves each with its own layer_image region; all of them lie in x 178.27..180.65, y 132.05..136.18 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L4PWR:hole 225.01,106.24|225.51,106.74'SLOT1.LVDS_2_N' crosses one gap in L4PWR's negative-plane field at 1 place, the longest 0.54 mm (fastest: SLOT1.LVDS_2_N at 100 ps 10-90 %, ASSUMED)
Along 'SLOT1.LVDS_2_N' on L5, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.54 mm under the track at (225.26, 106.61). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 225.01..225.51, y 106.24..106.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT1.LVDS_2_N' on L5 at (225.26, 106.61), 0.54 mm; other side: no plane on L6GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20225.01%2C106.24%7C225.51%2C106.74 serves each with its own layer_image region; all of them lie in x 224.26..226.45, y 105.49..107.83 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L4PWR:hole 225.01,107.24|225.51,107.74'SLOT1.LVDS_2_P' crosses one gap in L4PWR's negative-plane field at 1 place, the longest 0.5 mm (fastest: SLOT1.LVDS_2_P at 100 ps 10-90 %, ASSUMED)
Along 'SLOT1.LVDS_2_P' on L5, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.5 mm under the track at (225.26, 107.36). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 225.01..225.51, y 107.24..107.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT1.LVDS_2_P' on L5 at (225.26, 107.36), 0.5 mm; other side: no plane on L6GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20225.01%2C107.24%7C225.51%2C107.74 serves each with its own layer_image region; all of them lie in x 224.26..226.45, y 106.15..108.49 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L4PWR:hole 230.01,139.24|230.51,139.74'SLOT8.LVDS_10_P' crosses one gap in L4PWR's negative-plane field at 1 place, the longest 0.33 mm (fastest: SLOT8.LVDS_10_P at 100 ps 10-90 %, ASSUMED)
Along 'SLOT8.LVDS_10_P' on L5, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.33 mm under the track at (230.38, 139.49). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 230.01..230.51, y 139.24..139.74. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'SLOT8.LVDS_10_P' on L5 at (230.38, 139.49), 0.33 mm; other side: no plane on L6GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20230.01%2C139.24%7C230.51%2C139.74 serves each with its own layer_image region; all of them lie in x 229.26..231.58, y 138.46..140.49 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L9GND:hole 85.5,116.35|124.1,118.352 nets cross one gap in L9GND's negative-plane field at 2 places, the longest 8.12 mm (fastest: ETH_TXD1 at 1 ns 10-90 %, group)
Along 'ETH_TX_CLK' on L8, L9GND is a negative layer (copper everywhere inside the profile), and its field is cleared for 8.12 mm under the track at (119.44, 116.68). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. 2 places on 2 net(s) cross it, listed below, the longest first. What cleared it: the hole at x 85.5..124.1, y 116.35..118.35. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'ETH_TX_CLK' on L8 at (119.44, 116.68), 8.12 mm; other side: no plane on L7
- 'ETH_TXD1' on L8 at (120, 116.42), 6.45 mm; other side: no plane on L7
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL9GND%3Ahole%2085.5%2C116.35%7C124.1%2C118.35 serves each with its own layer_image region; all of them lie in x 115.2..125.18, y 115.35..117.68 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
IC22 TPS62125 (P2V5, 2.5 V) runs with a 10 uH inductor, which TI's recommended LC table does not list for 1.8-3.3 V outputs (15 or 22 uH only); loop stability is not covered by the datasheet (sheet 27)
IC22 TPS62125, which makes the DDR4 VPP rail P2V5 at 2.5 V, uses L2 = 10 uH. TI's recommended LC table lists only 15 uH and 22 uH for outputs from 1.8 V to 3.3 V. The output filter is therefore outside the combinations the datasheet supports for stability.
Evidence
Item Value Source Output voltage 0.8 V x (1 + 210k / (200k // 200k)) = 2.48 V sheet 27: R308, R40, R212 Inductor L2 VLCF5020T-100M1R1-1, 10 uH sheet 27; BOM Output capacitance (nominal) C297 22 uF 0603 X5R + 7 x 4.7 uF 0402 X5R + 7 x 100 nF, about 55 uF netlist P2V5; BOM codes Recommended LC, Vout 1.8-3.3 V 15 uH or 22 uH with 10 / 2x10 / 22 / 47 uF; no 10 uH row TPS62125 datasheet p.16, Table 2 Same part, IC32 (P5V_REG, 5.0 V) 10 uH with 22 uF, a listed combination sheet 27; Table 2 Why it matters
TI states that the TPS62125 "is optimized to work within a range of L and C combinations" and gives Table 2 for selection. A smaller inductance at this output voltage shifts the LC double pole and raises ripple current. TI's table gives no assurance for it, and the extra 4.7 uF capacitors at the DDR4 devices push the total past the largest value in the table. CERN's prototypes run, so this is a margin and documentation point rather than a known failure.
Possible fix
Use the 15 uH or 22 uH value from Table 2, or confirm the 10 uH / 55 uF combination against TI application note SLVA515, which Table 2 refers to for other combinations.
IC23 TPS51200 (DDR4 VTT) EN is pulled up to P3V3_REG while its VIN is P3V3 from another regulator, so EN can sit above VIN; TI says EN must stay at or below VIN at all times (sheet 27)
IC23 TPS51200 has EN (pin 7) on P3V3_PG, which R225 10k pulls up to P3V3_REG (IC37). Its VIN (pin 10) is P3V3 (IC31/IC41). TI says the EN voltage must remain at or below VIN at all times. On V5, EN goes to P3V3_REG as soon as IC31 releases PG_A and stays there until PG_A pulls low again, regardless of where P3V3 is within its power-good window.
Evidence
Item Value Source IC23 EN (pin 7) P3V3_PG netlist; sheet 27 P3V3_PG pull-up R225 10k to P3V3_REG (IC37 MIC5219-3.3, on whenever P5V_REG is) netlist; sheet 27 P3V3_PG driver IC31 PG_A (pin 26), open drain netlist; IRPS5401 datasheet p.8 IC23 VIN (pin 10) P3V3 (IC31 loop A + IC41) netlist Rule "Ensure that the EN pin voltage remains lower than or equal to VIN at all times." TPS51200 datasheet p.11, 6.3.5 (repeated p.25) EN absolute maximum 6.5 V, not tied to VIN TPS51200 datasheet p.4 Why it matters
P3V3_PG asserts once P3V3 is inside IC31's power-good window, not at its final value. It deasserts only when P3V3 falls out of that window, and P3V3_REG stays up as long as 12 V is present. EN therefore exceeds VIN briefly on every ramp-up and ramp-down, and whenever P3V3_REG reads higher than P3V3. The excess is small and within EN's absolute maximum, so no damage is implied. It is still a departure from a stated condition, and it is unchanged from V4 (R4 OBS-32).
Possible fix
Pull P3V3_PG up to P3V3 instead of P3V3_REG. IC22's EN, on the same net, is only limited to VIN + 0.3 V with VIN = 12 V, so it is unaffected.
IC3, IC25, IC37 (MIC5219-3.3) have ceramic-only output capacitors; the datasheet asks for about 1 ohm ESR and warns ultra-low-ESR capacitors can cause oscillation (P3V3_USB, P3V3_MP, P3V3_REG)
All three MIC5219-3.3 LDOs on V5 have only X5R/X7R ceramic output capacitors: IC3 (P3V3_USB, sheet 4), IC25 (P3V3_MP) and IC37 (P3V3_REG, sheet 27). Microchip specifies an output-capacitor ESR of about 1 ohm and warns that ultra-low-ESR capacitors could cause oscillation or an under-damped transient. IC25 and IC37 also have 1 uF nominal, which is the datasheet floor, before any derating.
Output capacitors as built
LDO Rail Output capacitors (BOM code) BYP pin Source IC3 P3V3_USB C160 22 uF 6.3 V X5R 0603; 11 x 100 nF X7R 0402 pin 4 unconnected sheet 4; netlist IC25 P3V3_MP C387 1 uF 25 V X7R 0603 (C259 100 nF is not to GND) pin 4 unconnected sheet 27; netlist IC37 P3V3_REG C320 1 uF 25 V X7R 0603; C58, C347, C348 100 nF X7R pin 4 unconnected sheet 27; netlist Requirement
Item Value Source Minimum output capacitance, no C_BYP 1 uF (0.47 uF below 10 mA load) MIC5219 datasheet p.13, 4.3 ESR "about 1 ohm or less and a resonant frequency above 1 MHz" MIC5219 datasheet p.13, 4.3 Ceramic warning "Ultra-low ESR capacitors could cause oscillation and/or under-damped transient response" MIC5219 datasheet p.13, 4.3 Recommended types "Most tantalum or aluminum electrolytic capacitors are adequate" MIC5219 datasheet p.13 Why it matters
A ceramic MLCC has milliohm-level ESR, the 'ultra-low ESR' case the datasheet warns against. The typical applications on p.3 use a 2.2 uF tantalum. On IC25 and IC37, X7R tolerance (+/-10%) and temperature coefficient (+/-15%) take the 1 uF part below the 1 uF figure. Their loads are a few mA, under the datasheet's 10 mA allowance, so capacitance alone is not the problem; the ESR is.
IC37 now feeds the IRPS5401 I2C pull-up rail (P3V3_REG). It replaced the AP7370 in V5 to fix CERN issue #377, so an LDO that rings would land on the bus that issue was about. V4 had the same ESR question on IC3 (R4 OBS-31); V5 extends it to IC25 and IC37.
Possible fix
Add a small series resistance (about 0.5-1 ohm) in each output capacitor branch, or use a tantalum/polymer output capacitor, as the datasheet's examples do.
IC31 IRPS5401 MTP strap R217 (3.48 kohm) selects MTP segment +5, but the sheet 26 configuration note says IC31 uses segment 6 (IC30's 2.87 kohm / segment 4 agree)
The schematic note for IC31 (sheet 26) says "MTP segment selected: 6" with "MTP resistor 3480 ohms". In Infineon's Table 9, 3.48 kohm on the MTP pin selects segment +5; segment +6 needs 4.12 kohm. IC30's strap and note agree (2.87 kohm, segment 4), so the IC31 note is not a different numbering convention.
Evidence
Item Value Source IC31 MTP pin (54) strap R217 3k48 0.1% to GND, C295 10 nF netlist; sheet 26 IC31 note "MTP segment selected: 6 / MTP resistor 3480 ohms" sheet 26 IC30 MTP pin (54) strap R209 2k87 to GND; note "MTP segment selected: 4 / MTP resistor 2870 ohms" netlist; sheet 25 Table 9 2.87 kohm = +4, 3.48 kohm = +5, 4.12 kohm = +6 IRPS5401 datasheet p.25 ADDR_PROT straps IC30 R208 2.87 kohm = offset +4 (0x14 / 0x44); IC31 R216 3.48 kohm = offset +5 (0x15 / 0x45), matching both notes IRPS5401 datasheet p.27 Table 11; sheets 25-26 Why it matters
The device loads its working registers from the segment the strap selects (datasheet p.25, 9.3). Both IRPS5401s are shipped unprogrammed (sheet 25/26 note: "This is unprogrammed part and need prior configuration to work"). If IC31's image is written to segment 6 as the note says, IC31 boots from segment 5 and its outputs (P3V3, MGT_1V8, MGT_0V9, MGT_1V2, P1V1) come up with whatever segment 5 holds. If the programming flow writes the segment the strap selects, nothing goes wrong and only the note is wrong. Either way the two documents disagree. This is unchanged from V4 (R4 OBS-30).
Possible fix
Correct the sheet 26 note to "segment 5". Or, if segment 6 is intended, fit 4.12 kohm for R217.
IC52 PS-GTR supplies PS_MGTRAVCC and PS_MGTRAVTT are left unconnected; UG583 permits that for unused PS-GTR but notes the PS MGTs must be powered for FPD zeroization
V5 leaves the Zynq's PS-GTR supply balls PS_MGTRAVCC and PS_MGTRAVTT unconnected, along with every PS-GTR lane and reference clock, since the board uses no PS-GTR lanes. UG583 allows unused PS-GTR supplies to be left unconnected or grounded, but adds that "PS MGTs must be powered to perform zeroization of the FPD". If CERN's firmware or security configuration ever relies on FPD zeroization, it cannot complete on this board.
Evidence
Item Value Source PS_MGTRAVCC (3 balls), PS_MGTRAVTT (3 balls) no net ($NONE$) netlist; UG1075 names via the IC52 bank map PS_MGTRRX/TX, PS_MGTREFCLK, PS_MGTRREF no net netlist Unused PS-GTR rule "If all of the PS-GTR transceivers in the Quad are not used, the associated power pins can be left unconnected or tied to ground." UG583 p.219 Zeroization note "PS MGTs must be powered to perform zeroization of the FPD." UG583 p.219 (added to the PS-GTR checklist, revision history p.367) Nominal values if powered PS_MGTRAVCC 0.85 V, PS_MGTRAVTT 1.8 V DS925 Table 2, p.4 Why it matters
Normal booting does not need the PS-GTR rails, and CERN's prototypes boot. AMD's note matters only to a design that uses FPD zeroization, for example a secure-boot or tamper-response configuration that clears the full-power domain. On V5 that path is closed by the hardware. Both rails exist nearby (P0V85 at 0.85 V and P1V8_AUX/MGT_1V8 at 1.8 V), so powering them through UG583's recommended filters is a small change if zeroization is ever needed. Until then, this is a documentation point for whoever writes the security configuration.
PS DDR4 ECC device IC17 leaves its unused upper byte's UDM_n/UDBI_n (E2), UDQS_t (B7) and UDQS_c (A7) floating; UG583 requires pull-ups on UDM_n and UDQS_t and a pull-down on UDQS_c (sheet 9)
IC17 is the x16 DDR4 that carries the PS ECC byte on its lower byte. Its unused upper byte has three balls with no connection at all: UDM_n/UDBI_n (E2), UDQS_t (B7) and UDQS_c (A7). AMD's DDR4 rules for an unused upper byte say these must be terminated: UDM_n and UDQS_t pulled up to VDD with 4.7 kohm or less, and UDQS_c pulled down to VSS with 4.7 kohm or less.
Evidence
Item Value Source IC17 E2 (NF/UDM_n/UDBI_n) NetIC17_E2, no other pin netlist IC17 B7 (UDQS_t) / A7 (UDQS_c) NetIC17_B7 / NetIC17_A7, no other pin netlist IC17 DQ8-DQ15 each on its own single-pin net (allowed to float) netlist IC17 lower byte DQ0-7, LDQS, LDM to PS_DDR_DQ64-71, DQS8, DM8 (the ECC byte), as UG583 requires netlist; IC52 bank map Rule "UDM or UDM_n/UDBI_n terminate to VDD with a pull-up resistor <= 4.7 KOhm"; "UDQS or UDQS_t terminate to VDD with a pull-up resistor <= 4.7 KOhm"; "UDQS# or UDQS_c terminate to VSS with a pull-down resistor <= 4.7 KOhm" UG583 p.45, DDR4 unused upper byte Ball functions E2 NF/UDM_n/UDBI_n, B7 UDQS_t, A7 UDQS_c MT40A1G16 ball map (stored pin facts) Why it matters
A floating strobe pair and data-mask input on a powered DDR4 leave the unused receivers at an undefined level. That is the condition AMD's rule exists to prevent, and it may add current or noise. CERN's prototypes boot with ECC enabled (wiki prototypes-evaluation), so this is a compliance gap rather than an observed failure. It is unchanged from V4 (R4 OBS-38, which named UDM only; UDQS_t/UDQS_c are in the same state).
Possible fix
Three resistors at IC17: about 4.7 kohm from E2 and B7 to P1V2, and from A7 to GND.
PS DDR4 ALERT_N is pulled up with 4.7 kohm to P1V2 (R113); UG583 asks for a 50 ohm pull-up to VDD on component interfaces (sheet 9)
DDR4_PS.ALERT_N joins IC52's PS_DDR_ALERT_N to the five DDR4 ALERT_n balls in fly-by and is pulled up only by R113, 4.7 kohm to P1V2. UG583 specifies a 50 ohm pull-up to VDD for Zynq UltraScale+ component interfaces.
Evidence
Item Value Source DDR4_PS.ALERT_N members IC52 PS_DDR_ALERT_N, IC13-IC17 ball P9, R113 netlist Pull-up R113 R0402_4K7 to P1V2 netlist; BOM Rule "terminate to VDD with a 50 ohm pull-up resistor" UG583 p.48 (alert_n) Why it matters
ALERT_n is an open-drain output the DRAMs pulse low to report CA-parity or write-CRC errors. With 4.7 kohm on a five-device fly-by net, the rising edge is about two orders of magnitude slower than with 50 ohm, so a short ALERT_n pulse may not return high cleanly at the controller. This only matters if parity or CRC is enabled in the PS DDR configuration. It is unchanged from V4 (R4 OBS-58).
Board over-temperature shutdown power-cycles instead of latching: the LM75s (IC36/IC39/IC40) whose O.S. outputs disable IC30 are powered from P3V3, which that shutdown removes (sheets 25, 28)
When any LM75 crosses its O.S. limit (80 C by default), its open-drain O.S. output pulls OVERTHERM_N, and through R252 0R the shared IC30 enable, low. That turns off IC30's rails, then IC31 and P3V3. The three LM75s run from P3V3, so they lose power, O.S. releases, R309 pulls the enable high again and the whole board restarts. A hot board therefore power-cycles about every conversion period rather than staying off until it has cooled.
The loop
``
mermaid sequenceDiagram participant LM as LM75 IC36/39/40 (+VS = P3V3) participant EN as NetIC30_28 (R309 10k to P5V_REG) participant P as IC30 -> IC31 -> P3V3 LM->>EN: O.S. low above Tos (80 C default) EN->>P: IC30 off, P1V8AUX_PG low, IC31 off P->>LM: P3V3 collapses, LM75 unpowered LM->>EN: O.S. released, R309 pulls EN high EN->>P: rails restart, LM75 re-powers with default limits LM->>EN: first conversion (up to 100 ms) still hot: O.S. low again``Evidence
Item Value Source O.S. of IC36, IC39, IC40 (pin 3) OVERTHERM_N, through R252 0R to NetIC30_28 = IC30 EN_L/A/B/C/D netlist; sheets 25, 28 Enable pull-up R309 10k to P5V_REG (always on with 12 V) netlist LM75 supply (+VS, pin 8) P3V3 (IC31 loop A, enabled by IC30's P1V8AUX_PG) netlist O.S. type and ratings open drain; O.S. output voltage 6.5 V absolute maximum, independent of +VS LM75A datasheet p.3 Power-up defaults Tos 80 C, Thyst 75 C LM75A datasheet p.4 Conversion time 100 ms LM75A datasheet p.4 Why it matters
The cut-out still limits temperature on average, but it does so by repeatedly hard-resetting the MPSoC and cycling every rail. Any Tos/Thyst values firmware wrote are lost at each cycle, because the LM75 restarts with its 80/75 C defaults. Holding the shutdown needs the sensors, or a latch on OVERTHERM_N, to stay powered from a rail the shutdown does not remove, such as P3V3_REG or P5V_REG. This is unchanged from V4 (R4 OBS-27).
WR clock DAC selects WR_DAC.SYNC1/SYNC2 (IC11/IC10 AD5662) float until the PL is configured: no pull-ups, and PUDC_B is strapped high so pre-configuration pull-ups are off (sheets 7, 13)
The two AD5662 DACs that set the White Rabbit VCXO control voltages (IC10 on SYNC2, IC11 on SYNC1) have their active-low frame selects driven only by PL pins IC52 F4/F5 in HD bank 94. Nothing pulls those pins up. PUDC_B (AH17) is tied high through R63, so the device's weak pre-configuration pull-ups are disabled. From power-up until the PL is configured, SYNC1, SYNC2, SCLK and DIN all float.
Evidence
Item Value Source WR_DAC.SYNC1 IC11 pin 5 (SYNC), IC52 F5, test pad B21; no resistor netlist; sheet 7 WR_DAC.SYNC2 IC10 pin 5 (SYNC), IC52 F4, test pad B18; no resistor netlist; sheet 7 WR_DAC.SCLK / WR_DAC.DIN IC10/IC11 pins 6/7, IC52 E5/E4; no resistor netlist PUDC_B (AH17) R63 1k to P1V8_AUX (high) netlist; sheet 13 PUDC_B = 1 "Weak preconfiguration I/O pull-up resistors disabled" UG1075 p.23 DAC outputs PLLDAC_OUT1/2 to the VCXO control inputs of OSC3 and OSC2 sheet 7 Why it matters
The AD5662 starts a frame on SYNC's falling edge and latches 24 SCLK bits. With SYNC, SCLK and DIN all floating, noise can clock a partial or complete word, or a power-down command, into either DAC before the PL takes over. The likely consequence is a VCXO pulled to an arbitrary frequency, or a DAC left powered down, until the White Rabbit gateware rewrites both DACs. That is benign if the gateware always initialises them, but the state is undefined. This is unchanged from V4 (R4 OBS-37).
Possible fix
Add 10 kohm pull-ups from WR_DAC.SYNC1 and WR_DAC.SYNC2 to P3V3_CLK, or ground PUDC_B so pre-configuration pull-ups apply to the whole PL.
Schematic title blocks: sheet 24 (Power-Up Sequence) has no revision and no checker, and sheet 10 (DDR-PL) has no check date; the other 28 sheets are complete
On V5, 28 of the 30 sheets carry a complete title block: document EDA-04514-V5-0, 'Sheet n of 30', revision, designer, drawer, checker and dates. Sheet 24 shows '-' for both revision and checker, and sheet 10 shows '-' where the check date belongs.
Evidence
Sheet Field As drawn Source 24 (power-up-sequence.SchDoc) Rev - sheet 24 title block 24 Check.by - (Designer and Drawn by: B.Guncic, 08.01.2025) sheet 24 title block 10 (ddr4-pl.SchDoc) check date - (other sheets: 10.11.2019) sheet 10 title block 1-30 Sheet n of 30, EDA-04514-V5-0 consistent on every sheet all title blocks V5 fixed the 'of 29' sheet count R4 reported (R4 OBS-52). Sheet 24 is the only sheet with no checker recorded, and it holds CERN's power-up sequence diagram, which several findings on this board rely on.
Sheet 11: the I2C_EEPROM harness port is labelled 'I2C_EEPROM.SCL', so the MAC EEPROM's data line compiles as net 'I2C_EEPROM.SCL.SDA' (also on sheet 6)
On sheet 11 the harness feeding IC8 (24AA025E48 MAC EEPROM) is labelled 'I2C_EEPROM.SCL' instead of 'I2C_EEPROM'. Its SDA entry therefore compiles as 'I2C_EEPROM.SCL.SDA' and its SCL entry as 'I2C_EEPROM.SCL'. The data line carries 'SCL' in its name on sheets 6 and 11 and in the netlist. The wiring is correct, so only the name misleads.
Evidence
Item Value Source Harness port label 'I2C_EEPROM.SCL' (compare 'I2C_WRFLASH' just above it) sheet 11 Data net I2C_EEPROM.SCL.SDA: IC6 pin 15 (SD5), IC8 pin 5 (SDA), IC35 pin 4 (SDA1), R147 netlist; sheets 6, 11 Clock net I2C_EEPROM.SCL: IC6 pin 16 (SC5), IC8 pin 6 (SCL), IC35 pin 3 (SCL1), R146 netlist Unchanged from V4 (R4 OBS-50). Renaming the port to 'I2C_EEPROM' gives I2C_EEPROM.SDA and I2C_EEPROM.SCL.
No local fiducials beside the 0.4 mm-pitch parts (IC30/IC31 IRPS5401 on top, IC19 LAN8831 on the bottom); assembly relies on three global fiducials per side
V5 has three global fiducials on each side (top FTG1/FTG8/FTG10, bottom FTG2/FTG3/FTG6) and no local fiducials. The finest-pitch parts are the two IRPS5401 PMICs (56-pin 7x7 mm QFN, 0.4 mm pitch) on top and the LAN8831 PHY (0.4 mm pad pitch) on the bottom; none has a fiducial pair beside it.
Evidence
Item Value Source Fiducials FTG10, FTG8, FTG1 (top); FTG3, FTG2, FTG6 (bottom); 1 mm pad, 2 mm mask opening board geometry (fiducial read) IC30, IC31 IRPS5401M, 0.4 mm pitch QFN, top IRPS5401 datasheet p.1; pad positions IC19 LAN8831, 0.4 mm pad pitch, bottom pad positions Whether local fiducials are needed at 0.4 mm depends on the assembler's process; many lines place 0.4 mm QFNs from global fiducials and part vision alone. Recorded so the assembler can confirm. Unchanged from V4 (R4 OBS-54).
PL DDR4 RESET_n (IC9 P1) is driven from 1.8 V bank 71 through a 100/240 ohm divider (R6/R302), so its high level is about 1.27 V, above Micron's RESET_n VIH maximum of VDD = 1.2 V (sheet 22)
The PL DDR4 interface sits in HP bank 65 at VCCO = P1V2, except its reset. DDR4_PL.RST_N_T leaves IC52 ball L22 (IO_T1U_N12_71) in HP bank 71, whose VCCO is P1V8. R6 (100 ohm) and R302 (240 ohm to GND) divide it down to DDR4_PL.RST_N at IC9 ball P1 (RESET_n). With the pin driven high at 1.8 V, RESET_n sits at 1.8 x 240 / 340 = 1.27 V. Micron's RESET_n input table gives VDD as the maximum high level, and VDD here is P1V2, 1.2 V nominal.
Evidence
Item Value Source Driver IC52 L22, IO_T1U_N12_71, bank 71 HP, VCCO = P1V8 (1.8 V) IC52 bank map (UG1075 names); netlist Divider R6 100 ohm 1% (L22 to RESET_n), R302 240 ohm 1% (RESET_n to GND) sheet 22; BOM RESET_n high level 1.8 V x 240 / (100 + 240) = 1.27 V (about 1.25-1.29 V with 1% parts and +/-1% on P1V8) calculation DRAM supply IC9 VDD/VDDQ = P1V2, 1.2 V nominal netlist Limit Table 84, RESET_n Input Levels (CMOS): VIH(AC)_RESET and VIH(DC)_RESET maximum = VDD MT40A1G16 datasheet p.258 Absolute maximum Voltage on any pin relative to VSS: 1.5 V MT40A1G16 datasheet p.253 Low during power-up R302 holds RESET_n low while L22 is unconfigured, which is the job UG583 gives the 4.7 kohm pull-down UG583 p.47 Why it matters
The pull-down side of the divider does what UG583 asks. The high level, however, sits about 70 mV above VDD for as long as the DRAM is out of reset, which is all the time in operation. That is inside the 1.5 V absolute maximum, so no damage is expected, but it is outside the recommended input range Micron specifies for RESET_n. It also costs about 5.3 mA of static drive from L22 (1.8 V / 340 ohm). The PS DDR4 (IC13-IC17) does not share this, because its reset comes from bank 504 at 1.2 V.
Possible fix
Change R6 to about 180 ohm, which gives 1.8 x 240 / 420 = 1.03 V. That is above 0.8 x VDD (0.96 V) and below VDD. Alternatively, drive the reset from a pin in a 1.2 V bank, such as bank 65.
FMC J1: PG_M2C (F1) is left unconnected, so the carrier has no 10 kohm pull-up on it and never reads it, against VITA 57.1 Rule 5.78; PG_C2M also reads 'good' by default until the PL drives it
VITA 57.1 requires the carrier to put a 10 kohm pull-up to 3P3V on both power-good signals, PG_M2C and PG_C2M. On V5, J1 contact F1 (PG_M2C) has no net at all. There is no pull-up and nothing reads it, so the carrier cannot tell whether an HPC mezzanine's own VREF/VIO_B supplies are good. PG_C2M (D1) has its pull-up (R327 10k to P3V3), but it is driven through T15 from IC52 B38. R328 holds T15's gate low, so PG_C2M sits high, signalling 'power good', whenever P3V3 is up and the PL has not driven B38. Rule 5.79's 'drive PG_C2M low when VADJ, 3P3V, 12P0V or 3P3AUX are out of specification' is therefore left entirely to gateware.
Evidence
Item Value Source J1 F1 (PG_M2C) no net ($NONE$): no pull-up, no receiver netlist J1 D1 (PG_C2M) FMC_PG_C2M: R327 10k to P3V3, T15 BSS138L drain netlist T15 gate FMC_PG_C2M_N: IC52 B38 (IO_T3U_N12_69, HP bank 69 at 1.8 V), R328 10k to GND netlist; IC52 bank map Rule 5.78 The carrier card shall provide a 10K pull-up resister on each of PG_M2C and PG_C2M signals to 3P3V VITA 57.1 p.68 Rule 5.79 A carrier card shall drive PG_C2M low when VADJ, 3P3V, 12P0V and 3P3AUX do not meet specification VITA 57.1 p.68 Rule 5.80 An HPC mezzanine drives PG_M2C low while VREF_A_M2C, VREF_B_M2C, VIO_B_M2C are out of specification VITA 57.1 p.68 T15 threshold VGS(th) 0.80-1.50 V, driven from a 1.8 V bank BSS138L p.2 Why it matters
A mezzanine that uses PG_M2C as an open-drain output, as Rule 5.80 allows, gets no pull-up from this carrier. The signal floats, and the carrier has no input to sample it. Of the supplies PG_M2C reports, this carrier uses VREF_A_M2C, which feeds bank 69's VREF ball, so it cannot tell whether that reference is good. On the PG_C2M side, the default is 'good' from the moment P3V3 rises, before any check of VADJ (P1V8) or 12P0V. Correct behaviour depends on gateware driving B38 high whenever a rail is out of specification. Nothing on the board gives the PL a power-good input for those rails, apart from reading the IRPS5401s over PMBus. The B38 drive is also only 1.8 V against T15's 1.5 V maximum threshold, which is enough for a 10 kohm load but leaves little gate overdrive.
Possible fix
Fit a 10 kohm pull-up from J1 F1 to P3V3 and route PG_M2C to a 3.3 V-tolerant input (an HD bank or PS MIO). For PG_C2M, consider gating it from the IRPS power-good outputs (P1V8AUX_PG/P3V3_PG) in hardware, so that it is 'not good' by default.
C442 (P0V85 bulk) is a Panasonic SP-Cap GX part, a series Panasonic says must not be used in accelerator or other continuous high-load applications; it points those to the long-life J/K/T series
C442, the only bulk capacitor on the 0.85 V MPSoC core rail P0V85, is an EEFGX0D471R: 470 uF, 2.0 V, from Panasonic's SP-Cap GX/GX-L series. Panasonic's datasheet states that GX/GX-L products must not be used for accelerator applications or other continuous high-load applications, and directs those to the long-life J/K/T series. The DIOT system board is a crate controller meant to run continuously, and P0V85 carries the ZU17EG core current through the 30 A TDA21535 stage.
Evidence
Item Value Source Part C442 EEFGX0D471R, 470 uF, 2.0 V, ESR 3 mohm max, ripple 10.2 A rms (100 kHz, 45 C) BOM; GX datasheet p.2 Restriction GX/GX-L must not be used for accelerator applications or other continuous high-load applications; use J/K/T series GX datasheet p.1 Endurance 2000 h at +105 C with rated voltage applied GX datasheet p.1 Ripple derating x0.7 between 45 and 85 C, x0.25 between 85 and 105 C GX datasheet p.2 Rail P0V85, 0.85 V (IC30 loop A through IC29 TDA21535, L9 PA2607.181NL); 1202 uF fitted in total netlist; rail clue Switching frequency IRPS5401 programmable 200 kHz-2 MHz; the loop A value is in the IC30 MTP image (IC30_signed.txt), unread IRPS5401 p.14, p.24 What is and is not at issue
The voltage fraction is comfortable: 0.85 V on a 2.0 V part is 43%. Ripple current depends on the loop A switching frequency held in the MTP image. At the 200 kHz floor with a low-tolerance inductor, the inductor ripple could approach the derated 7.1 A rms rating if C442 carried most of it. At typical frequencies it would be well below. The datasheet gives no rush-current limit. The open question is the one Panasonic states outright: whether continuous operation in an accelerator facility is the duty this series excludes.
Possible fix
Use the same footprint in a J/K/T long-life SP-Cap of equal or higher voltage and capacitance, as Panasonic recommends for continuous high-load use. Otherwise, record why the GX restriction does not apply to this board's duty.
PRST_N and PWRBTN_N carry the PS UART by default (IC48 drives PRST_N push-pull), while the crate spec puts push-button inputs on those Aux D-SUB pins
By default this board routes UART_PS to the backplane's PRST_N and PWRBTN_N lines: IC48 (SN74AVC2T245) drives TX push-pull at 3.3 V onto P1 H2 (PRST_N) and receives RX on P1 C3 (PWRBTN_N). In the crate those two lines go only to the Front Interface's Aux D-SUB15 pins 14 and 13, which the DI/OT Crate spec names PRST_N and PWRBTN_N, a push-button reset and power button. A reset button wired to the spec shorts the TX driver to GND, and neither button function exists with this board.
The path (OHWR-DIOT-CRATE-V3, both builds)
``
mermaid graph LR IC48["IC48 B1 (TX, DIR1 = P3V3, output)"] --> R392["R392 0R (fitted); R390 to the MPSoC not fitted"] --> P1H2["J14 (P1) H2 = PRST_N"] --> BP["Backplane J1.H2 → J8.A6"] --> FI["Front Interface J3.A6 → Aux D-SUB J5 pin 14"] FI2["Aux D-SUB J5 pin 13"] --> BP2["Backplane J8.A4 → J1.C3"] --> P1C3["J14 (P1) C3 = PWRBTN_N"] --> R393["R393 0R (fitted); R391 not fitted"] --> IC48b["IC48 B2 (RX, DIR2 = GND, input)"]``The facts
Point Value Source Routing "By default the UART_PS is connected to the backplane -> AUX conn, when USB is plugged in, UART_PS is connected to the FTDI" schematic sheet 4 note Fitting R392, R393 0R fitted; R390, R391 0R (to the MPSoC PRST_N/PWRBTN_N inputs) not fitted BOM / netlist TX driver IC48 SN74AVC2T245, DIR1 on P3V3 (A→B), OE low through T19 whenever no USB is plugged in (R388 10 k pull-up) netlist Crate assignment Aux D-SUB15 pin 13 PWRBTN_N, pin 14 PRST_N DI/OT Crate spec v3.1, Table 6, p.10 Other loads in the crate none: on the backplane PRST_N and PWRBTN_N run only from the system slot to the Front Interface OHWR0084 R3 netlist Why it matters
A user following Table 6 wires buttons to GND on pins 13 and 14. On pin 14 that holds an AVC output in a short to GND for as long as the button is pressed. Neither the reset nor the power button reaches the MPSoC, and nothing on the Front Interface or in the spec says the Aux connector carries a console UART. Either the spec and the front panel labels should say so, or the variant meant for the crate should fit R390/R391 instead of R392/R393.
DDR4 VPP (P2V5) is gated behind P3V3_PG while VDD (P1V2) has no interlock: the VPP-first order rests on IRPS5401 timing settings (CERN #384 workaround), not on the wiring
On V5, DDR4 VPP (P2V5, IC22) can only start after P3V3 power-good, but nothing in the wiring holds DDR4 VDD/VDDQ (P1V2, IC30 loop C) back until VPP is up. Whether VPP leads VDD therefore depends entirely on the IRPS5401 timing settings. CERN reached the recommended order on V5 by configuration (diot-sb-zu #384); the board's own sequence drawing (sheet 24) still shows VDD first, and if P3V3 never reaches power-good the DRAMs sit with VDD and no VPP.
Sequence as wired
- P1V2 is IC30 loop C. IC30's enables EN_L/A/B/C/D share NetIC30_28, so P1V2's start is set only by its TON_DELAY (sheet 25).
- IC31 is enabled by P1V8AUX_PG (IC30 PG_B). Its PG_A drives P3V3_PG once P3V3 is up (sheet 26).
- IC22 TPS62125 EN (pin 3) is on P3V3_PG, so P2V5 only starts after P3V3 is good (sheet 27; netlist IC22.3).
- CERN's power-up diagram (sheet 24) draws P1V2 ramping before P3V3_PG and P2V5 after it.
CERN's measurement
[diot-sb-zu #384](https://gitlab.com/ohwr/project/diot-sb-zu/-/issues/384) "DDR power sequence" (2026-04, closed): the datasheet order is 2V5 → 1V2 → 0V6, the configuration then gave 1V2 → 2V5/0V6. CERN's note: "As a work around, reducing 3V3_PG threshold (to get 2V5 up earlier) and increasing IC30 1V2 channel's TON_DELAY, it is possible to get recommended DDR sequence", with a scope capture. The IRPS5401 configuration files are not on this server, so Probe cannot confirm which settings a given board carries.
Requirement
Item Source VPP "must ramp at the same time or up to 10 minutes prior to" VDD; VPP at or above VDD at all times MT40A1G16 datasheet p.37 "Apply VPP without any slope reversal before or at the same time as VDD" MT40A1G16 datasheet p.38 What remains
- The order is a timing race between IC30's 1V2 TON_DELAY and the P1V8AUX_PG → P3V3 → P3V3_PG → IC22 chain, so its margin depends on load and temperature as well as on the configuration.
- If IC31 or IC41 fails to reach power-good, VPP never comes and VDD stays applied.
- Sheet 24 documents the old order.
Possible fix
Enable IC22 from P5V_REG or from IC30's enable net (NetIC30_28), so P2V5 rises before or with P1V2 by construction. P2V5 only supplies VPP, so starting it earlier has no other effect.
Five PESD3V3L5UY ESD arrays (D2, D6, D7, D8, D13) return to ground through a single via each
Five PESD3V3L5UY ESD arrays on the backplane, rear-I/O and JTAG lines (D2, D6, D7, D8, D13) each reach the ground plane through a single via, so that via's inductance sits in series with every clamp event; the datasheet asks for the transient return path to ground to be minimised and for ground vias.
What was measured
GND vias within 3 mm of each protector's ground pin (pin 2), from the drill layer stored for this take (via centre to pin centre). All five are on the bottom side; the nearest plane is L13GND, the layer next to BOTTOM_LAYER.
Protector Protects lines to GND vias within 3 mm of pin 2 D2 J14 (backplane P1) one, 0.81 mm D6 J33 (backplane P2) one, 0.75 mm D7 J14, J33 one, 0.81 mm D8 J33 one, 1.02 mm D13 J7 (JTAG header), J1 (FMC) one, 0.80 mm Contrast on the same board: D5 (same part, J14) has two vias (0.81 and 1.47 mm), D14 (J7) has seven within 3 mm, and D10 (PRTR5V0U2X, USB J2) has eight.
Why it matters
A clamp protects only as well as its path to the connector's return. The connectors return through their ground contacts into the planes, so each of these arrays reaches that return through one via. Nexperia's guidelines for this part (datasheet p10) ask to minimise the length of the transient return path to ground and, on multilayer boards, to use ground vias. A second via at each ground pad, as already done at D5, D10 and D14, halves that inductance.
The plane is one dielectric below the bottom layer here, so the via is short; whether one via is acceptable at the ESD levels the system needs is the owner's call.
- DecapViaFanoutminorOpen
DecapViaFanout: 2 components×2
Checked by the review, stands · Claude walker (Cern session) — Checked: the fanout length is read from the copper and is as stated; the 0.5 nH limit is Probe's starting value, not a datasheet or UG583 requirement. Whether it matters depends on this capacitor's role in the rail's PDN target. Not refutable with a citation; left open for the owner.
At C41, C44
Each finding, with its own headline
C41C41 on 'P0V85': its rail pad reaches its via over 2.43 mm of track, adding about 0.57 nH to its loop, past Probe's 0.5 nH starting value; put the vias at the pads (a 10 uF bulk capacitor: its inductance matters only far above the band it serves)
C41 on 'P0V85': its rail pad reaches its via over 2.43 mm of track, adding about 0.57 nH to its loop, past Probe's 0.5 nH starting value; put the vias at the pads (a 10 uF bulk capacitor: its inductance matters only far above the band it serves). C41.2 (rail) runs 2.43 mm along its copper to its nearest via, 1.43 mm past the 1 mm beside the pad: 0.57 nH; C41.1 (ground) not measured: the pad sits in a pour of its net on its own layer, which has no track length to measure. Each run past the pad as one conductor of a pair 0.92 mm apart (the pad pitch), 0.2 l ln(D/R) nH with R 0.127 mm (the track's width is not read): 0.57 nH in all. Put the vias at the pads (or in them), one per pad at least, the rail's and the ground's close together; the capacitor's own inductance as mounted is 0.55 nH (pad 0.07 nH + no via of the first net within 1 mm of its pad: the study's 0.6 nH per mm of 0.30 mm depth, ASSUMED). 0.5 nH is Probe's starting value, about the whole mounted inductance of a well-mounted 0402.
C44C44 on 'P1V8_AUX': its rail pad reaches its via over 2.39 mm of track, adding about 0.88 nH to its loop, past Probe's 0.5 nH starting value; put the vias at the pads (a 100 uF bulk capacitor: its inductance matters only far above the band it serves)
C44 on 'P1V8_AUX': its rail pad reaches its via over 2.39 mm of track, adding about 0.88 nH to its loop, past Probe's 0.5 nH starting value; put the vias at the pads (a 100 uF bulk capacitor: its inductance matters only far above the band it serves). C44.2 (rail) runs 2.39 mm along its copper to its nearest via, 1.39 mm past the 1 mm beside the pad: 0.88 nH; C44.1 (ground) not measured: the pad sits in a pour of its net on its own layer, which has no track length to measure. Each run past the pad as one conductor of a pair 3.00 mm apart (the pad pitch), 0.2 l ln(D/R) nH with R 0.127 mm (the track's width is not read): 0.88 nH in all. Put the vias at the pads (or in them), one per pad at least, the rail's and the ground's close together; the capacitor's own inductance as mounted is 1.79 nH (pad 0.20 nH + no via of the first net within 1 mm of its pad: the study's 0.6 nH per mm of 1.32 mm depth, ASSUMED). 0.5 nH is Probe's starting value, about the whole mounted inductance of a well-mounted 0402.
- ConnectorDomainMixedminorOpen
J14 takes 2 different supplies off the board on one connector: P12V0 (12.0 V), P5V0_MP (5.0 V)
J14 carries 2 different supplies out of the board on one connector: P12V0 (12.0 V) on A1/D1/E1/G1/H1/J1/K1, P5V0_MP (5.0 V) on B1. Declare the intent. Two rails on one connector is ordinary in a docking or power-distribution interface and a hazard in a signal one, and the difference is not visible from the netlist — but note that adjacent contacts at different voltages are exactly what a mirrored, reversed or off-by-one cable shorts together, so the cost of getting the mating wrong is a rail collision rather than a dead signal. If this is a standard connector, declare its family (connector_family_declare) and file the standard's contact map: supplies on the contacts it assigns to power are then not reported.
At J14
- CrystalLoadCapsminorOpen
Q2 has no load capacitors; IC2 (SI5341B-D-GM) integrates 8 pF - the crystal's CL is not on file to compare
Q2 (crystal): no load capacitor found on its terminal net(s) NetIC2_8, NetIC2_9, XTAL_SH. IC2 (SI5341B-D-GM) states its oscillator presents 8 pF (its datasheet p.30, single-read; XTAL mode; internal crystal load capacitors (CL) enabled; crystal 24.97-54.06 MHz), so the caps are integrated. Whether that matches the crystal cannot be said yet: Q2's load capacitance is not on file — and the datasheet on file for 7M-48.000MAHV does not name that exact ordering code, so it describes the family and a load capacitance read from it would be the family's range, not this part's. State it on the BOM row as parametrics.loadCapacitanceF (farads) from the part's own record, or submit parametric 'c_load' for 7M-48.000MAHV with a cite.
At Q2
- ReferencePlaneDiscontinuityminorOpen
ReferencePlaneDiscontinuity: 5 boards×5
5 checked by the review by Claude walker V1 (cern-01 session) — 4 different reasons
At gap:L2GND:hole 122.87,154.72|124.98,156.88, gap:L2GND:hole 172.23,132.3|172.43,136.05, gap:L2GND:hole 172.23,135.85|173.15,136.77, gap:L4PWR:hole 123.1,117.35|125.1,156.25, gap:L4PWR:hole 220.53,93.33|221.03,93.83
Each finding, with its own headline
gap:L2GND:hole 122.87,154.72|124.98,156.88'I2C_EEPROM.SCL.SDA' crosses one gap in L2GND and L4PWR's negative-plane field at 1 place, the longest 0.9 mm (fastest: I2C_EEPROM.SCL.SDA at 10 ns 10-90 %, group)
Along 'I2C_EEPROM.SCL.SDA' on L3, L2GND is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.9 mm under the track at (124.45, 156.65). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. 2 of them run ALONG the bare copper, parallel to its edges, rather than across it (the longest 0.49 mm). What cleared it: the hole at x 122.87..124.98, y 154.72..156.88. At 1 place(s) the track's plane is cleared on both sides of it, the same stretch on each reference layer: it has no plane on either side there, counted once. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- one stretch, cleared on 2 reference layers: 'I2C_EEPROM.SCL.SDA' on L3 at (124.45, 156.65), 0.9 mm, running ALONG the bare copper for 0.49 mm, not across it; other side: no plane on L4PWR / 'I2C_EEPROM.SCL.SDA' on L3 at (124.45, 156.65), 0.9 mm, running ALONG the bare copper for 0.49 mm, not across it; other side: no plane on L2GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL2GND%3Ahole%20122.87%2C154.72%7C124.98%2C156.88 serves each with its own layer_image region; all of them lie in x 123.16..125.79, y 155.31..157.94 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L2GND:hole 172.23,132.3|172.43,136.05'Main_DCXO_SDA' crosses one gap in L2GND's negative-plane field at 1 place, the longest 0.28 mm (fastest: Main_DCXO_SDA at 10 ns 10-90 %, group)
Along 'Main_DCXO_SDA' on TOP_LAYER, L2GND is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.28 mm under the track at (172.33, 135.79). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 172.23..172.43, y 132.3..136.05. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'Main_DCXO_SDA' on TOP_LAYER at (172.33, 135.79), 0.28 mm; other side: none: the signal layer is outermost
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL2GND%3Ahole%20172.23%2C132.3%7C172.43%2C136.05 serves each with its own layer_image region; all of them lie in x 171.23..173.43, y 134.69..136.89 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L2GND:hole 172.23,135.85|173.15,136.77'Main_DCXO_SDA' crosses one gap in L2GND's negative-plane field at 1 place, the longest 0.45 mm (fastest: Main_DCXO_SDA at 10 ns 10-90 %, group)
Along 'Main_DCXO_SDA' on TOP_LAYER, L2GND is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.45 mm under the track at (172.52, 136.14). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 172.23..173.15, y 135.85..136.77. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'Main_DCXO_SDA' on TOP_LAYER at (172.52, 136.14), 0.45 mm; other side: none: the signal layer is outermost
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL2GND%3Ahole%20172.23%2C135.85%7C173.15%2C136.77 serves each with its own layer_image region; all of them lie in x 171.46..173.58, y 134.94..137.34 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L4PWR:hole 123.1,117.35|125.1,156.25'JTAG_TCK' crosses one gap in L4PWR and L6GND's negative-plane field at 1 place, the longest 18.23 mm (fastest: JTAG_TCK at 2 ns 10-90 %, suggested by its name)
Along 'JTAG_TCK' on L5, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 18.23 mm under the track at (124.93, 146.95). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. 2 of them run ALONG the bare copper, parallel to its edges, rather than across it (the longest 17.34 mm). What cleared it: the hole at x 123.1..125.1, y 117.35..156.25. At 1 place(s) the track's plane is cleared on both sides of it, the same stretch on each reference layer: it has no plane on either side there, counted once. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- one stretch, cleared on 2 reference layers: 'JTAG_TCK' on L5 at (124.93, 146.95), 18.23 mm, running ALONG the bare copper for 17.34 mm, not across it; other side: no plane on L6GND / 'JTAG_TCK' on L5 at (124.93, 146.95), 18.23 mm, running ALONG the bare copper for 17.34 mm, not across it; other side: no plane on L4PWR
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20123.1%2C117.35%7C125.1%2C156.25 serves each with its own layer_image region; all of them lie in x 123.93..126.1, y 136.9..157.06 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
gap:L4PWR:hole 220.53,93.33|221.03,93.83'I2C_IRPS5401.SCL' crosses one gap in L4PWR's negative-plane field at 1 place, the longest 0.5 mm (fastest: I2C_IRPS5401.SCL at 10 ns 10-90 %, group)
Along 'I2C_IRPS5401.SCL' on L3, L4PWR is a negative layer (copper everywhere inside the profile), and its field is cleared for 0.5 mm under the track at (220.71, 93.73). The layer does not state the nets either side, so whether this is a split between two planes or a void in one is not said; either way the return current has to go round it. What cleared it: the hole at x 220.53..221.03, y 93.33..93.83. One decision covers the void: whether these tracks may cross it, or the plane should be filled under them. No crossing has a plane on the track's other side, so the whole return meets the break.
- 'I2C_IRPS5401.SCL' on L3 at (220.71, 93.73), 0.5 mm; other side: no plane on L2GND
Every site is a row of this finding's evidence, keyed by net, layers and position. GET /api/odb/boards/86/reference-planes?group=gap%3AL4PWR%3Ahole%20220.53%2C93.33%7C221.03%2C93.83 serves each with its own layer_image region; all of them lie in x 219.59..221.78, y 92.58..95 mm (board frame, Y up).
Read from the copper as drawn on the reference layer, along each track's centreline at exact breakpoints. The reference is the copper beside the track, never the layer's type or name. Whether it matters depends on each signal's edge rate, which nothing on file states for these nets: a slow static line crosses a split harmlessly, a fast edge does not.
9 clock/high-speed nets cross a plane gap with no return path, and RGMII lines on L8 run broadside over 16 off-board LVDS/FMC lines on L7
Nine clock, LVDS and DAC-control nets cross a gap in their reference plane with no plane on the track's other side, and the RGMII bus on L8 runs directly over 16 backplane-LVDS and FMC LA lines on L7 with no plane between: the return current has no continuous path at those sites, and the coupled noise leaves the board through J1, J14 and J33.
Read from the copper as drawn (Probe's reference-plane, plane-edge and off-board coupling screens on this take). Edge rates for these nets are not on file: the RGMII figures below use Probe's group starting value (1 ns 10-90 %, 3.3 V), which is not a sourced number, so the noise estimates are an order of magnitude, not a prediction. The geometry is exact.
Plane gaps under clock, LVDS and DAC-control tracks
Net Kind Layer Position (mm) Gap crossed Other side SLOT1.LVDS_2_N backplane LVDS (IC52.A22 to J14.C4) L5 over L4PWR (225.26, 106.61) 0.54 mm clearance, x 225.01-225.51, y 106.24-106.74 no plane on L6GND SLOT1.LVDS_2_P backplane LVDS L5 over L4PWR (225.26, 107.36) 0.50 mm clearance, x 225.01-225.51, y 107.24-107.74 no plane on L6GND SLOT8.LVDS_10_P backplane LVDS L5 over L4PWR (230.38, 139.49) 0.33 mm clearance, x 230.01-230.51, y 139.24-139.74 no plane on L6GND SLOT4.LVDS_0_P backplane LVDS L12 over L11PWR (233.35, 110.40) 0.56 mm clearance, x 233.01-233.51, y 110.24-110.74 no plane on L13GND ETH_TX_CLK RGMII TX clock (IC19.33 / R362) L8 over L9GND (119.44, 116.68) 8.12 mm along the L9GND void x 85.5-124.1, y 116.35-118.35 no plane on L7 ETH_TX_CLK (same net, plane edge) L8, nearest plane P3V3 on L10 (116.53, 116.68) P3V3 edge 0.31 mm (1.6H, H 0.2 mm) from the centreline for 7.39 mm, 18 % of the return beyond the edge no plane JTAG_TCK JTAG clock (J7.6, IC26, D13) L5 between L4PWR and L6GND (124.93, 146.95) runs ALONG the L4PWR void x 123.1-125.1, y 117.35-156.25 for 17.34 mm (18.23 mm affected), cleared on L6GND too no plane either side WR_DAC.DIN AD5662 data (IC10/IC11, from IC52.E4) L12 over L11PWR (189.79, 119.29) 0.39 mm clearance, x 189.45-189.95, y 119.13-119.63 no plane on L13GND WR_DAC.SYNC1 AD5662 IC11 SYNC (IC52.F5) L12 over L11PWR (189.93, 121.75) 0.41 mm clearance, x 189.68-190.18, y 121.38-121.88 no plane on L13GND WR_DAC.SYNC1 (same net) BOTTOM over L13GND (179.33, 134.14) 2.16 mm across the L13GND void x 172.36-179.78, y 131.93-136.50 outer layer WR_DAC.SCLK AD5662 clock (IC10/IC11, IC52.E5) BOTTOM over L13GND (179.65, 134.12) 2.02 mm across the same void outer layer ETH_TXD1 (RGMII data) crosses the same L9GND void for 6.45 mm at (120, 116.42) and runs 0.57 mm (2.9H) from the P3V3 edge on L10 at (117.25, 116.42); it is listed here as context. The voids on L9GND (y 116.35-118.35), L13GND (x 85.48-124.1, y 155.25-157.25) and L4PWR (x 123.1-125.1, y 117.35-156.25) line up as sides of one rectangle around x 85-125, y 116-157, which suggests a deliberate partition; whether these tracks were meant to cross or follow it is the designer's to say.
RGMII on L8 broadside over off-board lines on L7
L7 and L8 are adjacent signal layers with no plane between them where these runs sit. Probe's estimate is near-end noise at the starting edge above.
Off-board line (leaves at) Aggressor(s) on L8, coupled length, estimate FMC.LA_P0 (J1) ETH_TXD1 37.3 mm 65.2 mV; ETH_TXD0 29.2 mm 15.4 mV; ETH_TX_CLK 33.7 mm 6.3 mV FMC.LA_N0 (J1) ETH_RXD2 5.7 mm 13.6 mV; ETH_TXD1 34.9 mm 13.4 mV; ETH_TX_CLK 33.7 mm 10.6 mV; ETH_TXD0 29.2 mm 3.8 mV FMC.LA_N25 (J1) ETH_TXD3 37.6 mm 83.8 mV; ETH_TXD0 36.1 mm 43 mV; ETH_TX_CTL 24.7 mm 5.7 mV; ETH_RX_CTL 3.7 mm 4.6 mV FMC.LA_P25 (J1) ETH_TX_CTL 25.1 mm 34 mV; ETH_TXD0 36.2 mm 33.6 mV; ETH_TXD3 38.5 mm 19.1 mV FMC.LA_N14 (J1) ETH_TXD1 13.2 mm 29.1 mV; ETH_TX_CLK 23.8 mm 13.8 mV; ETH_TX_CTL 7.6 mm 9.8 mV FMC.LA_P14 (J1) ETH_TX_CLK 24.5 mm 19.6 mV; ETH_TXD2 6.9 mm 10.6 mV; ETH_TXD1 17.1 mm 8.3 mV FMC.LA_N11 (J1) ETH_TX_CLK 5.1 mm 14.6 mV FMC.LA_P11 (J1) ETH_TX_CLK 13.1 mm 10.7 mV; ETH_TXD1 6.1 mm 4 mV SLOT1.LVDS_3_N (J14) ETH_TX_CTL 3.3 mm 14.8 mV; ETH_RXD0 3.2 mm 13.6 mV; ETH_RX_CLK 3.2 mm 5.2 mV; ETH_TXD3 4.5 mm 3.9 mV; ETH_TXD0 3.2 mm 3.8 mV SLOT1.LVDS_3_P (J14) ETH_RX_CLK 3.2 mm 14.7 mV; ETH_TXD3 4.5 mm 13 mV; ETH_TX_CTL 3.3 mm 5.5 mV; ETH_RXD0 3.2 mm 3.8 mV SLOT2.LVDS_2_P (J33) ETH_RXD2 5.4 mm 21.8 mV SLOT2.LVDS_2_N (J33) ETH_RX_CTL 5 mm 13.7 mV; ETH_RXD2 5.1 mm 7.4 mV SLOT4.LVDS_15_N (J33) ETH_TXD2 3.9 mm 13.7 mV SLOT4.LVDS_15_P (J33) ETH_TXD2 4.4 mm 6.7 mV SLOT5.LVDS_15_N (J33) ETH_RXD0 3.1 mm 10.7 mV SLOT5.LVDS_15_P (J33) ETH_RX_CTL 2.8 mm 9 mV Why it matters
At a gap with no plane on the other side, the return current of a fast edge has to detour round the gap, which adds loop inductance and radiates; along a void (JTAG_TCK) the track has no reference for 17 mm. For LVDS and FMC differential lines a single-ended aggressor directly above one leg couples unequally into the two legs. Whether either matters at this board's real edge rates is not settled by anything on file: the PHY (LAN8831) and the MPSoC bank drive settings have no IBIS model here. Filling the clearances under the tracks, moving the crossings to where a plane continues on the other side, or putting a plane between L7 and L8 under the RGMII bus closes them.
Main parts
| Part number | Maker | Description | Where |
|---|---|---|---|
| 10052824-101LF | AMPHENOL | 72 Contacts Header, 6 Columns, 4 Pairs per Column | J31 |
| 10052825-101LF | AMPHENOL | 72 Contacts Header, 6 Columns, 4 Pairs per Column | J14 |
| 10052838-101LF | AMPHENOL | 96 Contacts Header, 8 Columns, 4 Pairs per Column | J11 |
| 1571983-4 or GDH04S04 | TYCO | 4 Ways SPST DIP Switch | SW1 |
| 22-27-2031 | MOLEX | Connector 3 Male | J6 |
| 24AA64T-I/MS | MICROCHIP TECHNOLOGY | 1.7V-5.5V, 64K (8K x 8-bit) I2C Serial EEPROM | IC34 |
| 39-28-1043 | MOLEX | Connector 4 Male | J13 |
| 502570-0893 | MOLEX | Connector MicroSD 8 Pin | J10 |
| 571-0122F | DIALIGHT | LED x2 Green 1C2A-BOT_3C4A-TOP | LD3 |
| 742792040 | WURTH ELEKTRONIK | L7 | |
| 7M-12.000MAAJ-T | TXC CORPORATION | Crystal 4 pins 1=XTAL Control 2=GND3=XTAL 4=GND | Q3 |
| 7M-48.000MAHV | TXC CORPORATION | Crystal 4 pins 1=XTAL Control 2=GND3=XTAL 4=GND | Q2 |
| 87832-1420 | MOLEX | Connector 14 Male | J7 |
| 93LC56B-I/OT | MICROCHIP TECHNOLOGY | 2K Microwire Compatible Serial EEPROM | IC49 |
| ABS06-32.768kHz-9 | ABRACON | Quartz Crystal | Q1 |
| APT1608LZGCK | KINGBRIGHT | LED Green 1C 2A | LD6 |
| ARJM11D7-502-AB-EW2 | ABRACON | 1 Port, Green/Yellow LEDs, RJ45 1Gigabit Ethernet, MagJack Connector Module | J5 |
| ASP-134486-01 | SAMTEC | Connector VITA 57, 400 Female contacts in 10 row abcdefghj | J1 |
| BAT54C | FAIRCHILD SEMICONDUCTOR | Dual Common Cathode Schottky Diodes | D9 |
| CDCM61004RHB | TEXAS INSTRUMENTS | Four Output, Integrated VCO, Low-Jitter Clock Generator | IC38 |
| FNETHE025 | PERICOM | Oscillator 4 pins 1=OE 2=GND 3=OUT 4=VDD | OSC4 |
| FRETHE025 | PERICOM | 3.3V CMOS 25.000MHz Ethernet VCXO | OSC2 |
| FT4232H-56Q | FUTURE TECHNOLOGY DEVICES INTERNATIONAL (FTDI) | Quad High Speed USB to Multipurpose UART/MPSSE IC | IC46 |
| HTSW-102-23-L-S | SAMTEC | 2 Mini Wire-Wrap Contact Strip Header | W1 |
| HTSW-104-07-L-S-LL | SAMTEC | Connector 4 Male | J3 |
| IS21ES08GA-JCLI | INTEGRATED SILICON SOLUTION INC. (ISSI) | 8GB eMMC with eMMC 5.0 Interface, 32GB NAND Flash | IC24 |
| LAN8831/Q2A | MICROCHIP TECHNOLOGY | Gigabit Ethernet Transceiver with GMII/MII/RGMII | IC19 |
| LP5951MF-3.0 | TEXAS INSTRUMENTS | 150mA Micropower Low-Dropout CMOS Voltage Regulator | IC7 |
| LTC4311CSC6 | TEXAS INSTRUMENTS | Low Voltage I2C/SMBus Accelerator | IC42 |
| MAX16025TE | MAXIM | Dual-Voltage, Capacitor-Adjustable, Sequencing/Supervisory Circuit | IC27 |
| MAX6639AEE | MAXIM | 2-Channel Temperature Monitor with Dual, Automatic, PWM Fan-Speed Controller | IC28 |
| MF-MSMF110/16 | BOURNS | Fuse | F1 |
| MT25QU01GBBB1EW9-0SIT | MICRON TECHNOLOGY | 1.8V, Multiple I/O Serial NOR Flash Memory, with Hold# Pins | IC18 |
| NDT3055L | FAIRCHILD SEMICONDUCTOR | N-CHANNEL MOSFET WITH DIODE AND 2 DRAIN | T1 |
| NX3225GA-25.000M-STD-CRG | NDK | Crystal 4 pins 1=XTAL Control 2=GND3=XTAL 4=GND | Q4 |
| PCA9548APW | NXP SEMICONDUCTORS | 8-Channel I2C-Bus Switch with Reset | IC6 |
| PRTR5V0U2X | NXP SEMICONDUCTORS | Ultra Low Capacitance Double Rail-to-Rail ESD Protection Diode | D10 |
| R0402_12K_1%_0.0625W_100PPM | GENERIC | Resistor - 1% | R344 |
| R0402_13K_1%_0.0625W_100PPM | GENERIC | Resistor - 1% | R178 |
| R0603_6K04_0.1%_0.063W_10PPM | GENERIC | Resistor - 0.1% | R47 |
About this design
- Design
- DIOT SB ZU — OHWR0083 R5
- Designed in
- Altium Designer
- Board
- 14 copper layers, 222.4 × 100 mm, 1104 components, 1208 nets
- Licence
- CERN-OHL-W-2.0
- Source
- https://edms.cern.ch/item/EDA-04514-V5-0/0 at commit
EDMS EDA-045 - Attribution
- CERN EDMS item EDA-04514-V5-0 'DIOT Zynq Ultrascale System Board - 16Gb DDR4', CERN EDA-04514 V5, variant V5-0 (16Gb DDR4): document 3399182 v.1, retrieved 2026-10-17T06:41:24Z. Copyright CERN 2022,2024-2026. Licence: CERN OHL-W v2.0+. Source location: https://gitlab.com/ohwr/project/diot-sb-zu
- Reviewed with
- Probe by E# (E-Sharp AB)