The LaST Upgrade Part 27 - STE RAM ISSUE

Last updated June 28, 2016

While doing various diagnostic tests on a STE motherboard I had, I noticed some odd RAM faults when running GB6. The faults only showed up during GEM WINDOW test, where when the green background was visible, I would see random white lines of a few pixels long appearing randomly on the screen. First I ram the Marpet Xtra RAM test software, it found no faults. I then tried the diagnostic cartridge , which showed no faults either. I suspected the DRAM but to cut a long story short, the STE worked fine with 1MB installed, but not 4MB. After a bit of investigation the problem got worse the longer the STE was turned on. Initially , up to about 2 hours, the machine ran perfectly, then progressively got worse.

Tests showed a problem with MAD9 on the DRAM circuit, though MAD8 also showed issues but they did not seem to cause any problems, at least not during the time I was testing it out.


The image above shows the noise problem on MAD9.



The image above shows a small fix I tried where changing to 33R resistors generally made things worse, but with 68R and 100pF capacitor the problem was greatly reduced but not cured.


After letting the STE "warm up" for some hours I finally got the errors as shown on the left. After adding the 100pF capacitor on MAD9, I got the result on the right. Clearly this was a lot better.


Unfortunately this solution was short lived :( It seems it would pass for 10 cycles, then get worse until it was failing almost every cycle. So it is clear that this problem gets worse the longer the STE is turned on. The odd thing is, after the STE has been turned on for several hours, MAD9 seems to have cured itself, but MAD8 still looks bad and the RAM test is still failing. I can only assume at this point that as the chip heats up internally, it changes the timings very slightly and "shifts" the fault around other address lines.

Adding more resistance or capacitance generally started making the problem worse. The extra resistance lowered the voltage to much, and the extra capacitance slowed down the signal until it was more like a triangle waveform.

I don't know how widespread this issue is. I am hoping some people can write some better RAM test software to diagnose this fault better.

I have also seen that this issue can be linked to DMA problems. If the RAM address being accessed isn't stable, then when the DMA access's RAM, then the address its working with would be incorrect. This could explain why some DMA issues seem to get worse the longer the machine is turned on. This also hints that 1MB machine may not suffer from DMA issues as much as 4MB machine. Of course this is only my initial thoughts.. More testing needs to be done.

So what's the cure for this ? Well the only cure is likely to be a buffer board for the DRAM circuit. While MAD8 & MAD9 seem to be the worst effected, all 9 address lines should be buffered. When I get some free time I will design a buffer board and then come back to this issue and see if I can fully solve it.