I spent a fair amount of time on the road teaching seminars on the effects of parasitic components (PC board leakages, stray capacitances, passive component imperfections, etc.) on analog circuits. As time went on and I began working on RF chips, package parasitics became a big part of determining chip performance, and getting the models right was pretty important. At low frequencies, the package is usually not considered important. At GHz frequencies, the package lead capacitances and bond wire inductances represent significant impedances and can impact the circuit.
A long time ago, in the first volume of Jim Williams' two-volume Analog Circuit Design series, Carl Nelson of Linear Tech said three things could kill a new chip: a new design, a new process, and a new package. He advised readers never to change more than one of those at a time. It was good advice that I didn't always follow. (In some cases, I wasn't allowed to follow it.)
A while back, we had a moderate-performance mixed-signal chip designed into a pretty good volume application. There was cost pressure as the volumes built up, so we undertook a die shrink to a smaller-geometry process. The supply voltage in the system was within the process breakdown range, and we thought the shrink would be OK. The customer felt the same way, since we had saved it some money. Low risk, same design, same package — just a teeny little shrink to the process to make the chip smaller and thus cheaper.
We tested the first samples of the shrunk version on the production tester. They performed fine on the bench and on the production tester. Everything went well until we shipped the new chips to the customer. Then the phone rang. The new chips didn't work right in the system. It took a while to sort it out, but one of the digital control signals on the customer's board going to our chip had noise on it. The original revision of the chip didn't respond to this noise, but the higher speed of the devices on the shrunk version propagated the noise into the power-on-reset circuit, and all the registers on chip were getting reset continuously.
Fortunately, I don't have a great anecdote about a package change causing a product problem. Most of that is due to the efforts of a really terrific packaging guy I worked with named Carl Roberts. We gave Roberts some real challenges over the years, like trying to find low-cost but reliable ways of plastic packaging multi-chip devices, switching to copper lead frame material for high-end audio chips, and figuring out how to reduce post-packaging stress on precision devices to preserve performance. He received a bunch of patents for his innovations.
Roberts became a fellow at Analog Devices for his contributions to the company's manufacturing technology; he was one of the few engineers outside the design community to gain that honor. While still at the peak of his game, he retired to pursue a second career as a deacon at his local church, where he did some really good work and helped a lot of people. Roberts died suddenly last year from a heart attack, and he is sorely missed by everyone who worked with him. Go hug your packaging guy now, before it's too late.