On Fri, Apr 9, 2021 at 12:33 PM Jon Steinhart <jon@fourwinds.com> wrote:
Lawrence Stewart writes:
> Regarding the SUN-1 design, I had heard a rumor that it was designed using TTL
> “typical” propagation delays rather than worst case, and as a result was fairly
> flakey.

It's astonishing how common a practice that was back then.

Even into the 2000s. I had a 6-month long war with one of the hardware guys for a time collection ISA card he did. It worked great, the driver worked great. Life was good. We shipped product. 5 years later, the customer comes back and wants a dozen more. So, we got new parts and 4 of the 6 new cards were flakey, 2 were good. Fingers pointed at the device driver, etc. Long months of intermittent troubleshooting continued for 5 months. During this time I build an ISA bus trace card, showed the traces were good and the flakiness was the result of bad data coming back from the card. At which point they brought in a different hardware guy to look at things. He discovered the first hardware guy had built an async circuit with typical delay patterns. One of the parts we used was rated at 200ns, but parts from the flakey board worked at 50ns. Turns out the manufacturer substituted a faster part, so the 'typical' delay propagation worked for this async circuit, but the faster response time would corrupt data from time to time. The design was tweaked to be synchronous with a latch, and the unmodified driver worked perfectly then...

Fun times that...

Warner