Sinclair ZX Spectrum 128 Troubleshooting Guide

This guide gives fault diagnosis for the Sinclair ZX Spectrum 128 "Toastrack". It combines the official Service Manual's test points and check sequences with the faults that surviving machines actually present with today.

Much of the 48K Spectrum's fault behaviour carries over, because the 128 shares its ULA-based video, its keyboard matrix and its DC-DC converter arrangement — so the Sinclair ZX Spectrum+ Troubleshooting Guide is a useful companion.

⚠️ Before you start

• There are no mains voltages inside the computer. It runs from an external supply; the highest rail on the board is +12 V. The mains transformer is inside the PSU brick.
• Never plug or unplug anything into the expansion edge connector with the machine switched on. On the Spectrum family this single habit accounts for the great majority of catastrophic faults — the connector shifts, shorts bus lines together, and takes out the power transistors and RAM.
• Identify your board version first (2, 2-1, 6K, 6U or 9G). Two documented defects are version-specific, and you will waste time chasing them as faults if you do not know which board you have.
• Hairline solder splatter can measure several ohms and produce very misleading symptoms — inspect visually before chasing anything complicated.

ZX Spectrum 128 circuit diagram, sheet 1 of 3, from the Service Manual. Sheets 2 and 3 are also on the wiki.

Known board defects that are not faults

Before diagnosing, rule these out — they are manufacturing defects, present from new:

Symptom	Affected boards	Cause and fix
Numeric keypad does not work at all	Issue 6K	An incorrect resistor value in the −Vcc supply to the MC1488 line driver (which serves both the serial and keypad ports) leaves the MC1488 without a proper supply. Replace R137 with a 15 Ω resistor. Corrected by Sinclair on the Issue 6U
A peripheral needing the bus clock does not work (e.g. DivMMC)	Versions 2 and 2-1 (Spanish/Investronica)	The /CLK line on the edge connector is not actually connected to the CPU clock. Add the link manually. Correct from Version 6K onwards
Beeper audio or sampled speech sounds wrong in some software	All 128s	Real 128 hardware contends banks 1, 3, 5 and 7 — the opposite of the published specification. Software written on real hardware sounds different on a +2A/+3, and vice versa. Not a fault
AY music quieter than beeper effects	All 128-family	The mix of the AY against the ULA beeper is poorly balanced by design. Normal behaviour

Normal power-on

A healthy machine shows the 128 menu — 128 BASIC, Calculator, 48 BASIC, Tape Loader, Tape Tester. Reaching this screen means most of the system is working.

Supply checks

Diagnose power before anything else. The on-board supply derives everything from the external unregulated input:

Rails, sources and loads

Rail	Derived from	Feeds
+5 V	7805 regulator	All logic
+12 V	TR4/TR5 inverter; rectified and smoothed by D15/C44	Video output stage
−12 V	Charge pump C111/C112 with D28/D29, off the same square wave	MC1488 RS-232 driver and the keypad
−5 V	From the −12 V rail via zener D19	Expansion port

The external pack should hold up under load: at minimum mains (215 V AC) and 1.4 A output the trough must not fall below 7.0 V; at maximum mains (265 V AC) and 600 mA the peak must stay below 13 V.

If the ±12 V rails are missing or low, the inverter is the suspect. This is the same circuit used on the 48K machines, where TR4, TR5 and the coil frequently fail together — replacing only one of them often results in immediate re-failure. With the converter working you should be able to hear the coil buzzing faintly.

On the 128 these rails feed the video stage and the RS-232/keypad drivers rather than the RAM, so a converter failure here shows up as video problems, a dead serial port or a dead keypad — not the 48K machine's white screen.

Intermittent resets, especially when the machine is moved, are usually the power plug or the power socket rather than anything on the board. Replace a loose socket; do not attempt to tighten it.

Clock checks

Expected oscilloscope readings (Service Manual)

Signal	Test point	Expected
Master crystal	IC37 pin 6	17.734475 MHz, no jitter
Z80 clock	IC2 pin 6	3.54689 MHz, no jitter
Colour encoder	IC1 pins 46, 47	8.8 MHz, no jitter
Sound carrier	IC38 pin 4	Within 2 kHz of 6.0 MHz (UK) or 5.5–5.9 MHz (European versions)

Source: Service Manual. If the sound carrier is out of tolerance it can be adjusted per the setting-up instructions rather than replaced. The crystal is a precision part — series resonant with 20 pF and specified to ±10 ppm absolute — so a machine with drifting colour or sound is worth checking here before suspecting the encoder.

If it does not initialise

With rails and clocks proven, work through the Service Manual's sequence:

1. Check +5 V and ground distribution to the ROM, Z80, ULA and RAM.
2. Check RD, WR, MREQ, D0–D7 and A0–A15 from the Z80 — all should be active immediately after a reset.
3. Check the RAS/CAS lines to the uncontended RAM (IC15–IC22) — active immediately after reset.
4. Check the RAS/CAS lines to the contended RAM (IC6–IC13). Press RESET to obtain a clear trace.
5. Confirm the ROM (IC5) is enabled by an active-low signal at pin 20.
6. Confirm the bank register loads correctly — immediately after reset, pins 2, 5, 7, 10, 12 and 15 should be low.
7. Check the outputs on the RGB connector.
8. Check the picture on a television and listen for key clicks on each ENTER press. Then check LUM0 at IC36 pin 7, the sound carrier at IC38 pin 4, and the drive into the modulator (the DC level at the bottom of that waveform is typically 185 mV).

Note that the Service Manual's own symptomatic fault table was never completed — it is printed as "TBD" with an author's note that details would follow once in-service history was known. The symptom tables below therefore draw on the 48K manual's equivalent section and on community repair experience.

RAM faults

RAM is the most common electronic failure on a surviving 128.

• All devices must be 4164, 150 ns, 128-row refresh. Mixing speed grades or fitting slower parts causes instability that looks like a logic fault.
• Shorted RAM pulls down the 5 V rail. Powering the board from a current-limited bench supply is the fastest way to find it: the board goes straight into limiting and the offending devices get hot. Remove the hottest first; it is common to find several.
• After the rails are proven, feel each RAM device. None should be warm. A warm RAM chip is a failing one.
• Diagnostic ROMs are far quicker than probing. Brendan Alford's ZX Diagnostics runs from the ROM socket and reports failing data bits with beeps and coloured bars even when the display RAM itself is faulty; Paul Farrow's 128K RAM tester then exercises all eight banks.
• Where several devices in a bank have failed, socketing and replacing the whole bank is usually faster and kinder to the board than chasing them one at a time.

The 48K machine's POKE-based RAM location test also works on a 128 running in 48 BASIC and will identify a failed device in the lower 16 K — see the Sinclair ZX Spectrum+ Troubleshooting Guide for the table.

Video faults

Vertical "jailbars" on RGB — a documented 128 fault

A distinctive Toastrack complaint: faint vertical banding on plain, light-coloured areas that appears on the RGB output but not over RF or composite. It is a noise/ground-plane artefact rather than a failing component, and it persists even after a routine recap.

The established fix on a 128 is:

• Replace C28 with a 47 µF axial capacitor. The original is 22 µF, located to the right of the 5-way keyboard connector.
• Replace C7 and C8 with 1 µF ceramic capacitors. These sit next to the IC12 and IC13 RAM devices at the bottom left of the board.

This significantly reduces the effect, though it does not eliminate it completely. The equivalent fix on a grey +2 is C52 instead.

Other video symptoms

Symptom	Action
No picture at all, machine otherwise runs	Check the RGB connector outputs and LUM0 at IC36 pin 7; check the modulator drive waveform
Picture but no colour	Check the 8.8 MHz colour-encoder clock at IC1 pins 46/47; then the TEA2000 (IC36) and its surrounding components
No RF picture, RGB fine	Modulator or the MC1376 driver (IC38); check the sound carrier is within tolerance and try retuning the television
Corrupt or striped display	RAM; run a diagnostic ROM
Wavy lines across the screen	Press each electrolytic in turn — if the lines clear while pressing one, that capacitor is the fault

Sources: Service Manual and the 48K field-service guide.

Sound faults

• No sound at all (neither beeper nor AY): suspect the ULA first — on the Spectrum family, faulty or missing sound most often identifies the ULA.
• Beeper works, AY silent: the AY-3-8912A (IC32) or its supply. Test with a 128 BASIC PLAY command rather than a beeper BEEP, which uses the ULA path.
• AY audible but much quieter than beeper effects: normal for all 128-family machines, not a fault.

Keyboard faults

The 128 uses the Spectrum+ keyboard assembly, connected by two ribbon tails at KB1 (5-way) and KB2 (8-way).

• A dead block of five keys points to the 8-way tail or its connector; a dead block of eight points to the 5-way tail. This follows the 48K matrix layout.
• Most keyboard faults are simply membrane failure — the pads wear out or the tail cracks with age. Clean and reseat both tails before suspecting anything electronic.
• Two Spectrum+ assembly quirks apply to the 128 as well: bottom-row keys failing can often be cured by removing the ribbon clamps, cleaning between the ribbon layers around the clamped area and adding packing when re-clamping; and a key that repeats endlessly is caused by the membrane backing plate being over-tightened — slacken the screw nearest that key by a quarter turn.
• When refitting the tails, make sure they are fully inserted and not kinked.

Serial port and keypad faults

Both the RS-232 port and the keypad run through the MC1488 driver (IC33) and MC1489 receiver (IC34), and both depend on the −12 V rail.

1. If both the serial port and the keypad are dead, check the −12 V rail first — the charge pump (C111/C112, D28/D29) and zener D19. A dead −12 V rail disables both at once.
2. If only the keypad is dead and the board is an Issue 6K, this is the R137 defect described above, not a fault you can chase — fit a 15 Ω resistor.
3. Otherwise check the connectors themselves (6-way BT-type jacks) and then substitute IC33 and IC34.

Overheating

The Toastrack's external heatsink exists because the 128 dissipates significantly more than a 48K machine.

• Check the heatsink is securely fastened and making good thermal contact; apply fresh thermal compound to the regulator when refitting.
• Inspect the board around the heatsink fixings for hairline cracks — on the Spectrum family these are a documented cause of intermittent faults.
• An intermittent that appears after a few minutes and clears on cooling, with nothing obviously hot, suggests a thermal-expansion connection fault — a cracked track, a dry joint or a tired socket — rather than a failing semiconductor.
• In normal running the ULA runs warm and the regulator hot; RAM should stay cool.

Contacts, sockets and random crashes

A large proportion of "random crash" faults on Spectrums are contact problems rather than component failures:

• Remove, clean and firmly reseat every socketed IC.
• Clean the expansion edge connector fingers with a soft eraser and contact cleaner.
• Re-tension the EAR and MIC socket contacts if tape loading is intermittent — push the spring contact down and inward rather than replacing the socket.
• Check the power socket; replace it if loose.

Common fault catalogue

• Keypad dead, everything else fine, Issue 6K board — the R137 defect; fit a 15 Ω resistor.
• Serial port and keypad both dead — the −12 V rail; check the charge pump and D19.
• DivMMC or clock-dependent peripheral not working, Spanish board — /CLK not connected on the edge connector; add the link.
• Vertical jailbars on RGB only — replace C28 with 47 µF and C7/C8 with 1 µF ceramics.
• Board draws excess current, chips hot — shorted RAM.
• Crashes, corrupt display — RAM; test with a diagnostic ROM.
• No video, no serial, but logic runs — the ±12 V inverter (TR4/TR5/coil).
• No sound at all — ULA. AY silent, beeper fine — AY-3-8912A.
• No colour — 8.8 MHz encoder clock, then the TEA2000.
• Dead block of keys — the corresponding membrane tail.
• Faults appear only when warm — heatsink mounting, hairline cracks, or the ULA.
• Random resets when moved — power plug or socket.
• Beeper audio sounds wrong in specific games — contention behaviour, not a fault.

Related pages

• Sinclair ZX Spectrum 128
• Sinclair ZX Spectrum 128 General Maintenance
• Sinclair ZX Spectrum 128 Capacitor Replacement Guide
• Sinclair ZX Spectrum+ Troubleshooting Guide — the 48K machine, whose DC-DC and keyboard behaviour the 128 shares
• Sinclair ZX Spectrum +2 Troubleshooting Guide — the Amstrad repackaging of this machine
• Sinclair ZX Spectrum Keyboard Membrane Repair