Bare CPU Printed Circuit Board For The Alpha NT XL366 Workstation I Designed Back In 1995 Or So. This

Bare CPU Printed Circuit Board for the Alpha NT XL366 workstation I designed back in 1995 or so. This was an obscure model of an obscure product line, made by a company (Digital Equipment Corp.) that is now itself obscure. To be honest I don't even remember much about this machine now.

What I do remember is the HUUUUGE fight I got into with our Signal Integrity team while I was designing this, over decoupling capacitors.

Decoupling caps are small components that hold a charge to help even out power when a circuit is active. This board featured hundreds of them, smaller than a grain of rice (see photo comparison of mounting pads vs rice grain below).

Our Signal Integrity team was tasked with making sure everything was electrically stable, so they required many hundreds of these to be added to the board, based on power simulations they did. Trouble was, they wanted so many, we couldn't even build the board.

My job as the Systems Engineer here was to meet the requirements from the SI team, but also from manufacturing, and the requirement that my PCB layout techs don't go insane trying to place and route the board. SI really only cared about signal quality, so they would not relent, and I ended up getting shouted at at one point by a junior SI engineer who was also under a lot of stress, when I said "There are different schools of thought on this.." and he screamed THERE ARE NOT DIFFERENT SCHOOLS OF THOUGHT ON THIS!!

It got to the point where the product was not going to get built, because we just couldn't fit like a thousand of these tiny caps on the board, we needed to ditch at least 25% of them to have a hope. The models were the models though, and you couldn't argue against them.

But then my boss got a genius idea. What if we could prove the simulation models were too conservative? We came up with an experiment where we would remove caps from an older system and measure the power supply noise, to see how many caps could be taken off before the system became unstable.

Me and the junior SI engineer were tasked with doing this experiment (later deemed The Decapitation Project), so we grabbed a Tektronix scope and Metcal soldering station and headed over to this abandoned lab we had in our old Maynard headquarters, a now creepy attic space on the 6th floor of an old mill building. Here were a few older Alphastation 3000 workstations we built years earlier, working but waiting to be recycled.

We had this special program that would thrash the CPU within an inch of its life, to put a big demand on the power supply system. While this was running, the SI engineer measured the power quality, while I proceeded to (very carefully to avoid short-circuiting the system) actually desolder caps from the board while the workstation was running.

We managed to get about 1/3 of them off before there was any noticeable effect, and we found one specific type of cap was not doing much of anything at all. We took the data back to the head of the SI team, and he finally relented and let us remove several hundred capacitors. (He also buried the report and data I had, because he didn't want the bad publicity - I remember being mad about that)

The system got built after that, and worked just fine. We did try to enact a small bit of petty revenge on the SI team manager though - there was a recognition event for people involved on the project, and me and our PCB procurement guy decided to give the SI team manager a special "Faraday Award" for achievement in capacitance (Farads are a measure of capacitance - geeky eng joke). We took an old bowling trophy with a giant, beer-can sized electrolytic capacitor strapped to the top of it as the award. He was a no-show so we didn't get to present it. Those SI guys never did have much of a sense of humor.

Anyway, long story sorry. Just thinking of it recently because I was helping someone at work with an analog simulation and I remembered this..