ETA-10

My recollection was that the theoretical speedup of 4x was what one GATE would do if it lived alone. However once you put all these gates together (ie, 20K of them), it was the long and short path routing around that die that ultimately limited the total DEVICE speedup. Further, DEVICE to DEVICE path lengths on the PCB limited the total system clock rate and we put a lot of stuff in there to dynamically tune the clock nets across the board but in the end, the highest reliability occured clocking at 7nS in liquid and 14nS in air. We did a lot of speed testing by shmooing the clock and bin'ing the parts based on how fast they would go at room temp. Ultimately, some statistics were developed that helped us further "guess" which ones would make it to at least 7nS in liquid. Some didn't though and you didn't find that out until they were on the board and the board was in the liquid. We got pretty good at not doing that near the end though. Accurate speed testing and looking at ring oscillator rates at room temp helped predict the performance in liquid pretty well as I remember it. Each device was pretty much capable of testing itself in this process. Using the BIST (boundary scan test logic) we could shift in an initial pattern, run the device at various clock rates for exact numbers of clock cycles and then shift off the results to compare against simulation. We'd do initial testing at 4 million clocks and then a more production like test at 4 billion clocks. The difference between the two was the cost (in waiting for the run) of simulating that many cycles on a CY205. Running the actual part that many cycles was a no brainer. We used various off the shelf GPIB pulse generators as the clock source in a bench tester and we had a ALSI20K array (AAZB to be specific) that handled this BIST interface on the ETA10 PCB and ultimately we put that array on a PC-AT plug-in card and it became part of our bench top tester. Was fun stuff to be sure. I am not aware of any other cryogenically cooled machines in that time frame. We were always competing against Cray doing liquid cooled (Fluorinert) systems and other air cooled Japanese designs at the time. Chris On Friday (02/14/2014 at 03:42AM +0000), Alexey Toptygin wrote: -- Chris Elmquist