Five Volts Away From a Very Expensive Paperweight

Retrocomputing Hardware Restoration 🎮 Play: Continuity

The smell got there before the diagnosis did. I plugged in the dollar VIC-20 from the estate-sale ribbon box — the same model I learned BASIC on in 1988 — expecting a blinking cursor and instead got that unmistakable warm-varnish reek of an electrolytic capacitor letting go. No cursor. No colour. Just a forty-three-year-old machine telling me, through my nose, that something inside had failed the way old power supplies fail.

So today is not about restoring anything. It is about one number: 5.0 volts, and why getting it wrong turns a rare machine into a board of dead ceramic.

Why the rail is a loaded gun

The logic in these machines — the 6502 CPU, the 6522 VIAs, the DRAM — is NMOS, and the 74-series glue is TTL. Both have an absolute-maximum supply voltage of about 7 V. Not a recommendation. A cliff. Feed the rail 9 V for a couple of seconds and the substrate junctions punch through; the chip does not reset, it dies, permanently, and half of them are no longer manufactured.

Now, how could 5 V ever become 9 V? Because most of these supplies regulate with a linear series-pass regulator — a 7805 on the VIC-20’s own board, or, on the C64, the notorious potted “black brick.” A 7805 works by acting as a variable resistor between input and output, burning off the difference as heat. When it fails, it usually fails the worst possible way: the pass transistor cooks and goes short-circuit. Input connects straight to output. The unregulated ~9 V upstream lands directly on your 5 V rail, and every chip downstream is now past its absolute maximum. This is the single most important fact in the hobby, and it is why the first commandment is never trust the original PSU.

The crowbar: deliberately blowing your own fuse

The community’s canonical defence — Ray Carlsen’s over-voltage protector, if you want the schematic by name — is a crowbar circuit. The name is literal: if the rail goes over-voltage, you throw a metaphorical crowbar across it, dead short, and let the fuse take the hit instead of the silicon.

Four parts, in-line on the +5 V rail:

  +5V rail ───[ FUSE 2A ]───┬──────────────► to logic board
                            │
                     anode ─┴─ SCR (e.g. C106 / 2N6504)
                            │  gate
                            ├───[ R 100Ω ]──┐
              6.2V zener ───┤               │
                            │             gate
                    ground ─┴───────────────┴─── ground

At 5 V the 6.2 V zener is dark — no current, the SCR sits open, everything passes. The instant the rail climbs past ~6.2 V, the zener conducts, dumps current into the SCR’s gate, and the SCR latches hard: anode to cathode becomes a near-short across the supply. Enormous current flows, the 2 A fuse opens in milliseconds, and the rail collapses to zero before it ever reaches the 7 V cliff. You lose a twenty-cent fuse and keep a hundred-dollar chip. The trip point is a genuine engineering compromise: set the zener too close to 5 V and normal ripple nuisance-trips you; too high and you clip the chips’ funeral a little late. 6.2 V threads it.

Build it on a scrap of perfboard, put it between supply and board, and test it without the computer attached first — feed the input from a bench supply, wind the voltage up slowly, and watch it fire around 6.2 V. Only once it has proven it clamps do you let it anywhere near the VIC.

While the board was open I did the archaeology, too. Every chip carries a four-digit YYWW date code, and the newest one dates the machine — the same read I was doing on spruce when I cored a tree for its rings, except the growth here is a factory’s, not a forest’s. Newest code on my board: 8341. Week 41 of 1983.

The reeking cap turned out to be a 1000 µF filter that had lost its electrolyte and half its capacitance; I’ll recap the whole cluster and reform nothing this old by shocking it — a slow voltage ramp on the bench first. The machine still doesn’t boot. But the rail now sits at a measured 5.02 V behind a fuse that will die before the 6502 does, which is more honest protection than Commodore ever shipped. When I broke a fifty-year-old typewriter drawband in April I decided restoration is just debugging with older parts. Turns out some of those parts are actively trying to kill the patient, and the first repair is teaching the supply not to.