oh nice 4MB 30pin with parity, and board supports it.

I just bought this stick off ebay. Optosys Topless Panther1 256MB SD-RAM 168-pin PC-133U Non-ECC CL2.

rasz_pl wrote on 2023-01-08, 20:16:

So those arent partials. Someone was saving on packaging costs like 30pin TOPLESS simms.

I had some of these once, with a big 'TOPLESS' sticker on them. I think it's the same people 😉

btw I found another company making simms from partially defective chips in early nineties from UK - Memory Corporation plc aka Torridon plc http://www.adamsmith.com/case_memorycorporation.html One of the first SSD patents was filed by Torridon, later acquired by Micron.

Of course the ultimate example of UK use of partially defective chips (in the early eighties) was Sinclair's computers. The Spectrum was specced to use 32kb chips, but frequently was built with half-defective 64kb chips instead. No wonder these machines were so much cheaper than competitors...

dionb wrote on 2023-09-27, 11:02:

Of course the ultimate example of UK use of partially defective chips (in the early eighties) was Sinclair's computers. The Spectrum was specced to use 32kb chips, but frequently was built with half-defective 64kb chips instead. No wonder these machines were so much cheaper than competitors...

Did "actual" 32kbit chips actually exists? Most of the DRAM chips have a square matrix of bits, with each side of the sequare being the same power of two, so the total number if bits is a power of 4. 16k and 64k are powers of four, whereas 32k is no (integer) power of four. I used to think that all "32kbit" chips sold that time were (possibly) half-defective 64kbit chips.

How about Centon CHP128M/MPSIMM 128MB EDO 60ns 72-Pin SIMM. Both Toshiba tc5165405bfts and NEC d42565405g5 are 16Mbit x4 3.3V EDO ram. What I dont quite understand are power regulators suggesting 5V supply when ram chips are not 5V tolerant, unless thats what 6x Pericom PI5C 3861Q '10-Bit, 2-Port Bus Switch' chips are for? Pass signals thru resistors one way, enable transparent connection the other way? Weird ram.

rasz_pl wrote on 2023-10-06, 16:27:

unless thats what 6x Pericom PI5C 3861Q '10-Bit, 2-Port Bus Switch' chips are for? Pass signals thru resistors one way, enable transparent connection the other way?

These kind of bus switch can in fact be used for voltage limitation. They use similar bus switches on late PCI graphics cards like the GeForce 5200 PCI. The GeForce chip is not 5V tolerant, just as the RAM chips on these modules, yet the GeForce 5200 PCI is on a PCB with the 5V voltage coding.

The idea is that these switches are basically nothing more than 10 enhancement-mode N-Channel FET switches. This means that the FET provides a low resistance between the two "busses" when the gate voltage is sufficiently higher than the bus voltage, but blocks conduction when the gate voltage is not at least ~0.7 to 1V higher than the bus voltage. If you supply these FET switches with around 4.0 to 4.3V, they are unable to drive the gate of the internal FET high enough to pass voltages exceeding around 3.3V. I found datasheets for similar FET switches that explicitly mentioned this applications. Some vendors even produce these FET switches with a silicon dropper diode in the Vcc line to drop 5V supply voltage to the required supply voltage to make them a 3.3V clamp.

This kind of clamp is definitely not a precision clamp, but the only really important property is that the leakage current at ~3.6V is sufficiently low that the ESD clamping diodes in the 3.3V RAM do not get overloaded, and they pass signals well enough up to 2.5V or something like that, to not inhibt 3.3V CMOS or 5V TTL high levels from propagating through them. It's noteworthy that the channel resistance of this bus switch is specified at 0V (5 ohm typical) and at 2.4V (10 ohm typical), but they don't mention the channel resistance at higher signal voltags - because it will be quite high, and passing 2.4V is good enough to pass a TTL high level.