I considered this as a possibility, and tried a MAX1232. It made no difference at all, i.e. the parts that worked with the Intel reset circuit continued to work, and the ones that didn't, continued to fail.

These "original" Intel (and Intel-licensed) NMOS/HMOS parts worked fine, to the satisfaction of the industry, for over a decade before "reset" IC's became popular. As Erik has suggested, the reset issue was not an issue before the days of FLASH program store.

I've tried these parts in a circuit that provides a reset of exactly 24 clock cycles (counting from an external oscillator that drives the MCU, after 64K counts from the application of Vcc > ~4.85 volts followed by the first oscillation at the oscillator input to the MCU. A crude comparator (LM311) senses the rise of Vcc against a reference zener, then gates on the oscillator to a 16-bit counter, the terminal count of which produces a reset pulse that's 25 clock cycles long, which is slightly longer than the required two instruction cycles long as specified, and only then is the RESET dropped.

That didn't work any better than the RC reset either. What's more, they start to run on power-up, and only the schmidt-triggered pushbutton reset causes the devices to fail.

Whether this means these devices are defective or not, I'm not planning to use them. I suspect they were routed to me from STC or one of the several disk drive makers up north of here, who didn't want to fool with 'em.

I suspect that these RESET issues are either FLASH-related, or ATMEL-related. I don't know whether Philips or other MFG's parts have reset issues. I won't have the ATMEL problem, and the Philips, Siemens, and Signetics parts I have in house function fine with a 10 uF cap to Vcc and an 8K2 pulldown. The Dallas parts have the reset and watchdog IC built in, so no external help is needed. Frankly, I doubt that so long a reset is needed except while the power/oscillator relationship settles. Every other MCU that I've used has done fine with a much shorter reset pulse.

RE