>From the last round of power-cycle testing, I was left with the suspicion that temperature had an effect on the frequency at which NOR corruption occurs. So I looked into controlling and monitoring the M1's temperature while power-cycling.
The idea was to set up the M1 in a room with the temperature maintained by air conditioning and to monitor the M1's temperature with one of my lab instruments. All this would be connected to my workstation PC with an USB-over-Cat5. Since me entering the room to look at LEDs to verify the system's state would have disturbed the setup, I also had to automate this part of the test process. The BOOTSTS register I discussed in a previous posting provides the necessary information. While working on automating the test loop, I set up temperature monitoring and checked the performance of the air conditioning at various settings. Unfortunately, there were several problems with this: first, my lab meter (Picotest M3500A), which was in the same room, apparently does not compensate for its own temperature. Since I used a thermocouple (from an U60T, probably K-type), this meant that the measurement results, while looking plausible just around 20 C, were in fact useless. Second, the USB-over-Cat5 adapter produced a high rate of errors, causing the USB bus to reset from time to time, and the USB host/host driver even downgrading the bus from high-speed to full-speed. The USB instability also meant that I couldn't usefully operate the M1 (with the JTAG board) at the end of this link. I therefore postponed further attempts to control the temperature. I had better luck with automating the test loop. The BOOTSTS test proved to be reliable. I also added automated data collection and recovery after a NOR failure. The loop could therefore run indefinitely without any human intervention. I also used a few suggestions from Sebastien to configure the M1 such that it would start rendering more quickly. This brought down the test cycle time from about 80 seconds to less than 20 seconds. Next: what happened in the 10516 test cycles that followed. - Werner _______________________________________________ http://lists.milkymist.org/listinfo.cgi/devel-milkymist.org IRC: #milkymist@Freenode
