Design review — control paths and I/O (sequencer logic/memory, IO, bus tester, backplane, video) — 2026-09-21¶
Section 1 of the 2026-09-21 control-paths and I/O review: the sequencer logic v2.1, findings 1.1-1.8. The other cards' sections ("above" and "below" in the original), the cross-card notes and the summary:
hardware/reviews/control-io-2026-09-21.md.
Second pass after the mechanical one in mechanical-2026-09-21.md. Read from the KiCad netlists (reports/netlist.net, pin
types from the Eagle symbols), chip values from the Eagle boards, the firmware that drives the ATmegas
(embedded/bus-tester/bus-driver, embedded/sequencer-card/sequencer4) and the microcode generator
(firmware/microcode/ucode-generator2). Gate pin-outs were checked against the symbol pin functions (I0/I1/O for
7400/08/32, O/I/I for 7402; 74175/74/192/244/374 pin functions as printed in the netlist).
Every finding: the trace (ref.pin -> net -> ref.pin, chip types), the failure scenario, severity (HIGH = stops the card / damages parts, MED = unreliable or latent, LOW = housekeeping), and how to confirm it on the bench. Already-known items (BACKLOG.md, video README) are not repeated except where a new detail changes them. Report only; no redesign.
Scripts used (not committed): a netlist dumper (parts + nets with pin function/type), an SV1/SV2 pin-by-pin comparator, a bus-driver-table-vs-tester-wiring check and a microcode-bit-vs-RAM-wiring check; the results of the three checks are in the "checked, no issue" lists below.
1. Sequencer logic v2.1 (hardware/cards/sequencer-logic/kicad/v2.1)¶
1.1 HIGH — 17 bus lines are driven by the logic card no matter what -BUS-EN says (extends the BACKLOG item)¶
The BACKLOG's "CPU off switch" item assumes that de-asserting -BUS-EN silences the card. It silences the nine pipeline 74LS374s (IC7, IC12, IC14, IC16, IC17, IC19, IC20, IC28) and the two microcode-address 74LS244s (IC15, IC35), and nothing else. These stay on the bus:
| Bus pin(s) | Driver | Enable / control | Trace |
|---|---|---|---|
| B9–B12 REG-RD-ID0..3, B13–B16 REG-LD-ID0..3 | IC4 74LS244 (both halves) | -ONE-OPERAND-SEL = IC31 74LS04 p4 | IC16 74LS374 p16 (7Q, -2-BYTE-OPERAND-SEL) -> IC31 p3; IC31 p4 -> IC4 p1/p19 and IC18 p19 |
| same 8 lines | IC5 74LS244 | -2-BYTE-OPERAND-SEL directly | IC16 p16 -> IC5 p1/p19 |
| C3–C6 ADDR-REG-ID0..3 | IC18 74LS244 B half (p9/7/5/3) | -ONE-OPERAND-SEL (IC18 p19) | inputs = LADDR-REG-ID0..3 from IC19 74LS374 |
| C3–C6 | IC18 A half, IC11 74LS244 A half | SRC-ADDR (IC17 p5 -> IC18 p1), DEST-ADDR (IC16 p19 -> IC11 p1) | inputs = operand register IC6 74LS374 |
| B18 -REG-LD-LO | IC31 74LS04 p10 (totem pole) | none | IC27 74LS08 p11 -> IC31 p11 |
| B20 -REG-LD-HI | IC31 74LS04 p12 (totem pole) | none | IC24 74LS08 p6 -> IC31 p13 |
| C30 -RESET | IC36 74LS04 p6 (totem pole) | none | front-panel RESET latch (IC32 74ALS02 p4) -> IC36 p5 |
| C27 OUT | IC22 74ALS02 p13 (totem pole) | none | OUT-ON/OUT-OFF SR latch IC22 gates C/D |
| C28 -BUS-EN | IC36 74LS04 p8 | none | known (BACKLOG) |
IC4 and IC5 have complementary enables (IC31 inverts one into the other), so one of them is always on. With -BUS-EN high, IC16 p16 floats, IC31 p3 reads high, IC31 p4 goes low and IC4 + IC18B drive whatever the floating LREG-/LADDR- nets happen to read onto B9–B16 and C3–C6.
Failure: the bus tester (bus-driver.ino setup()) makes every signal in its table a push-pull MCP23017 output —
including -RESET, OUT, -REG-LD-LO/HI, REG-RD-ID, REG-LD-ID, ADDR-REG-ID — and puts the address and data buses in
WRITE mode; only BR-COND, -INT and IN are left as inputs. With the logic card fitted there are two totem-pole
drivers on each of these 17 lines. The tester reads back wrong levels and, worse, an LS244/LS374 output high
(short-circuit current 40–225 mA) into an MCP23017 output low (absolute maximum 25 mA per pin) can kill the
expander pin. This is why the tester "cannot load RAM with the logic card fitted" — but the v2.2 fix as written
(OEs of the 374/244s through a switch, OC driver for -BUS-EN) would still leave B9–B20, C3–C6, C27 and C30 driven.
Confirm: logic card fitted, sequencer READY high, tester plugged in. With the CPU held in reset, drive REG-RD-ID0 (B9), -REG-LD-LO (B18), ADDR-REG-ID0 (C3), OUT (C27) and -RESET (C30) from the tester and read them back; a meter on the line shows an intermediate level (~0.5–1.5 V) instead of the commanded one.
1.2 HIGH (documentation / rebuild) — IC33 and IC34 are drawn and BOM'd as 74LS192 (BCD decade); the microcode addressing needs binary counters¶
Trace: N$4 (IC38 74LS08 p6) -> IC33 UP p5; IC33 QA p3 -> IC35 74LS244 p2 -> p18 = CNT-CLK (the pipeline
clock, via IC25 74LS32 p13/p11 = BUS-LATCH-CLK); IC33 QB/QC/QD (p2/6/7) -> IC35 -> CADDR0/1/2 (SV1 p14/16/18);
IC33 CO p12 -> IC34 UP p5; IC34 QA/QB/QC -> CADDR3/4/5; IC34 QD p7 -> COUNT-FAULT -> IC26 74LS32 p12. Each
microstep is two counter clocks (QA is the phase bit, QB.. are the step address); 64 steps need a 7-bit binary
count. The generator (controlLine.c writeCurrentLine, main.c loadNextInstruction) writes lines 0,1,2,… per
opcode and the fetch alone is lines 0–5.
With decade parts the step address sequence would be 0,1,2,3,4, 8,9,10,11,12, 16,… (line 5 of the fetch — the one
that drops -REG-UP and -MEM-RD — never executes, then line 8 runs) and COUNT-FAULT fires after 40 steps, not 64.
The machine executes the switch-ROM program (MACHINE.md 2026-09-21), so binary parts are almost certainly what is
soldered in — but Sequencer-Logic-Prod-V2.1l.brd (value 74LS192N), the BOM CSV and both KiCad symbols (74*192)
say decade. The same symbol/BOM entry appears 16 times on the index-register card (register/eagle/v1.1 BOM: 16 ×
74LS192N for the 16-bit registers, where decade counters would make the PC count in BCD) — outside this review's
scope but the same discrepancy. A rebuild from the BOM would order the wrong part.
Confirm: read the markings on IC33/IC34 (and on the register cards' counters). If they are 193s, correct the
value/BOM; if they really are 192s, the fetch sequence works only by accident of which lines are skipped and
tests/sequencer should dump the step address (JP4 p3 / SV1 CADDR pins) while single-stepping.
1.3 MED — The pipeline is not reloaded when reset is asserted, only when it is released; during reset (and from power-up until the first reset release) the 374s drive the last/random word onto the bus¶
Trace of the reset-release pulse: RESET (IC32 74ALS02 p4) -> IC37 74LS04 p1->p2 (N$44) -> IC37 p9->p8 (N$42) -> IC36 74LS04 p11->p10 (N$46) -> IC32 p9; IC32 p8 = RESET; IC32 p10 = N$32 = NOR(RESET, RESET delayed 3 gates). N$32 is a ~30–45 ns positive pulse only on the falling edge of RESET; it is ORed with CNT-CLK in IC25 (p12/p13 -> p11 = BUS-LATCH-CLK). COUNT-RESET (IC26 p3 = RESET OR (UCODE-COUNT-RESET AND N$4)) clears IC33/IC34 during reset, so CNT-CLK falls to 0 and never rises. Nothing clocks the pipeline while RESET is high; nothing clears it either (-RESET goes only to IC8/IC9 74LS175 CLR p1). -BUS-EN stays asserted (IC36 p8 follows READY).
Consequence: while the front-panel switch sits in RESET the bus carries whatever the pipeline last held — the step
that was executing when reset was hit, or power-up garbage. BACKLOG.md's description ("in reset the pipeline still
drives ... word instruction 0 step 0") is what happens after the switch goes to EXECUTE (N$32 latches address
IR=0/step=0); during reset it is worse than that. Memory-card side: -RESET presets the boot flip-flop (memory IC12
74LS74 PRE p4) so the 28C64 answers every address (FORCE-ROM); memory IC6 74LS04 p9->p8->p3->p4 (N$4) feeds
-MEM-WR straight to the -WE (p27) of IC1, IC2 and the EEPROM IC13; -ROM-CS comes from the address decode with -VMA
(IC7 74LS138 G2A). A stale pipeline word with -MEM-WR and -VMA asserted therefore writes the EEPROM during reset.
The ROM was byte-identical to rom/shipped on 2026-09-18, so either it has not happened yet or the fitted 28C64
has data protection.
Confirm: scope B24 (-MEM-WR) and C12 (-VMA) at power-up and with RESET held, several times; run
tools/verify_firmware.py against the chip after a batch of power cycles. Related: the N$32 pulse (3 × LS04 tpd)
is right at the LS374's 15 ns minimum clock width — fine with LS parts, marginal if IC36/IC37 are ever replaced
by faster families.
1.4 MED — SPARE3 pipeline bit is wired to the sequencer-memory ATmega reset through JP2, and the microcode holds that bit at 0¶
Trace: RAM IC5 I/O5 (SV1-side name SPARE3) -> memory JP1 p1 -> logic JP1 p1 (USPARE3) -> IC16 74LS374 p7 (3D);
IC16 p6 (3Q, SPARE3) -> JP2 p2; JP2 p1 -> -MEM-CPU-RESET -> logic JP1 p2 -> memory JP1 p2 -> ATmega IC15 p1
/RESET and all five MCP23017 /RESET (p18), R4 10k pull-up. The generator's initCurrentLine
(main.c:94) calls clearSignal on every name; for a name without a '-' prefix that is bitOff = 0
(main.c:76-86). SPARE3 is never set anywhere, so every microcode line has SPARE3 = 0.
Failure: if JP2 is ever fitted the sequencer-memory card is held in reset (READY low, -BUS-EN high, RAM control lines floating) and the CPU can never run. Harmless today only because JP2 is open. Confirm: JP2 open.
1.5 LOW — -REG-LD-LO / -REG-LD-HI are combinational with unequal paths (latent glitch)¶
-REG-LD-LO = NOT(LREG-LD-LO AND N$53) [IC27 74LS08 p13/p12 -> p11 -> IC31 p11 -> p10 -> B18];
N$53 = IC30 74LS32 p6 = (REG-LD-ID != 0) OR (branch latch N$71 AND REG-LD-ID == 0), where REG-LD-ID is taken from
the bus (B13–B16) through IC25/IC30 ORs, IC37 p13->p12, IC27 p9/p10; the branch latch N$71 is the gate SR latch
IC26 p8 / IC37 p3->4 / IC30 p3 / IC37 p5->6. Two gates on the LREG-LD-LO path versus seven on the ID path: if a
step raises LREG-LD-LO on the same edge that moves REG-LD-ID from non-zero to 0 with the branch not taken, B18
pulses low for ~40–60 ns before N$53 falls (a runt load strobe to R0 on the register cards' 74LS32 inputs
IC1/IC15/IC25/IC47 p10). The current generator does not do that: branch.c writes setLdId(PC) one line before
it raises REG-LD-LO/HI, and REG-LD-ID is 0 from initCurrentLine anyway. A future microcode that changes the ID
and the strobe together will hit it. Confirm (if the microcode is changed): scope IC31 p10 across the step edge.
1.6 LOW — Interrupt jumper JP3 needs two jumpers; edge mode triggers on the release of -INT¶
IC23A 74LS74: CLK p3 = INT-EDGE (JP3 p2), PRE p4 = IC36 p12 = NOT(INT-LEVEL, JP3 p4), CLR p1 = IC36 p4 = NOT(INT-START OR RESET), D p2 = VCC. -INT (C13) is JP3 p3, GND on p1 and p5, pull-up selectable by the -INT-PULLUP header (R5 10k). Mode select = -INT to p2 (edge) or p4 (level); the other input must be jumpered to its GND neighbour, otherwise IC36 p13 floats high (PRE permanently asserted, DO-INT whenever enabled) or the LS74 clock floats. In edge mode the LS74 clocks on the rising edge, i.e. an interrupt is registered when -INT is released, not when it is asserted. Confirm: JP3 population; in edge mode pulse -INT low from the tester and watch IC23 p5 set on the trailing edge.
1.7 LOW — Counter clock gated by asynchronous levels¶
N$4 = IC38 p6 = (UCODE-CLK AND RUN AND NOT(COUNT-FAULT OR DO-HALT)) with UCODE-CLK = (QG1 AND FREE-RUN) OR (SINGLE-STEP/WAIT AND SINGLE-STEP-CLK) (IC38 p8/p11, IC26 p6). RUN (IC32 p1), the halt flip-flop (IC23B p9) and the SS/WAIT latch (IC29) change asynchronously to QG1, so CONT, EXECUTE or flipping SS/WAIT while running can produce a runt pulse on IC33 UP. Front-panel only; the switch latches (IC29/IC32 NOR pairs with R6–R11 pull-downs) are debounced, the HALT/CONT push buttons (R3/R4 pull-ups into IC31) are not, but they only feed a set-only / clear-only flip-flop, so the bounce is harmless.
1.8 LOW — All pipeline outputs float when -BUS-EN is high (no pull-ups anywhere)¶
With READY low (54 s microcode copy) or the tester holding -BUS-EN high, IC16's -INT-JMP (p12) and -BRANCH-RD (p2) — the output enables of IC10/IC3 and IC21/IC13 (74LS374 on DATA0–15) — float, as do OPERAND-CLK, BRANCH-LD-, INT-LD-, OUT-ON/OFF, LD-INS-REG, UCODE-COUNT-RESET, CNT-CLK and COUNT-FAULT. LS inputs read a floating pin as high, which is the inactive level for all of them (COUNT-FAULT high even stops the counter clock), so it works by convention; noise on -BRANCH-RD/-INT-JMP could put the branch/interrupt registers on the data bus while the tester drives it. Confirm: with READY low, scope IC16 p2/p12.
Checked, no issue (sequencer logic)¶
- SV1/SV2: all 80 pins carry the same signal on both cards (U-prefix on the logic side); JP1 1:1 SPARE3 / -MEM-CPU-RESET / SRC-ADDR / DEST-ADDR on both cards.
- Instruction register IC8/IC9 74LS175: CLK = IC24 p3 = LD-INS-REG AND RUN, CLR = -RESET; opcode source is IC2 74LS244 (DATA0–7, enable DO-INT low) or IC1 74LS244 (all inputs VCC = $FF, enable -DO-INT) — complementary through IC36 p1->p2.
- Microcode address: CADDR6–13 = IC15 74LS244 from IC9/IC8 Q outputs; CADDR0–5 from the step counter (1.2).
- Operand paths: IC6 74LS374 (OPERAND-CLK, OC = GND) -> IC5/IC18A/IC11A; IC4 vs IC5 exactly one enabled.
- Branch latch and conditional PC load logic as the Notes.md V2.1 change list describes; BR-COND (C24) input only.
- OUT SR latch IC22 gates C/D; IN (C26) unconnected on this card; -RUN (C29) unconnected everywhere.
- Halt: IC23B set by (FP-HALT AND UCODE-COUNT-RESET) OR SOFT-HALT (IC24 p11, IC25 p6), cleared by RESET OR CONT (IC32 p13); SOFT-HALT and USOFT-HALT are one net (Notes.md item resolved).
- Spare gates: IC37 p11 grounded; IC11B inputs grounded and enabled (harmless).
- Step-clock source jumper SS-SEL (front-panel latch IC29 gates C/D or external JP4 p2); JP4 exposes UCODE-COUNT-RESET.
Last updated 26 October 2026