Anyone know where uPD2167 SRAMs appeared?

[allison]

On 10/09/2018 08:54 PM, Josh Dersch via cctalk wrote:
> On Tue, Oct 9, 2018 at 5:14 PM Ethan Dicks via cctalk <cctalk at classiccmp.org>
> wrote:
>
>> Hi, All,
>>
>> I asked a version of this question earlier this year.  I have not been
>> able to find any vintage machines that used these 16Kx1 55ns SRAMs.
>> Anyone recognize them?  Lots of them for sale on eBay.  Probably few
>> buyers.  One would want to know which systems used them, thus my
>> question.
The uPD2167(16Kx1) and 2147 (4kx4) rams appeared in 1981, I was at NEC then.
They were offered as 65/55/45ns and for the time that was faster than
most static and dynamic parts.

>> They probably would have been excellent in a DEC MOS memory board but
>> I have no evidence they were used thusly.  Contemporary DRAMs were
>> cheap and 64Kx1 so that's what was in consumer gear.
>>
>> Anyone?  Fast SRAM?  Anywhere?
>>
> The Three Rivers PERQ used 48 of them for microcode store in the 16K CPU,
> and on the Z80-based IO Processor.  (I suspect the IO Processor didn't need
> RAMs quite that fast, but 3RCC probably had a lot of them on hand due to
> their use in the main CPU...)
>
> - Josh

Popular use was memory planes for high speed graphics and cache for some
minis.
At that time and for a good while there were no micros that pushed the
speed needs
that hard even 12mhz 8086 in the mid 80s.
>> There's little point in wiring 8 of them up into a byte vs using a
>> 62256 except for speed.  55ns is faster than any 8MHz machine really
>> needs (100ns-150ns was typical for those depending on bus
>> architecture).  I could see these being cache RAM for a minicomputer
>> vs primary RAM.
>>
>> -ethan
>>

I used them for a fast cull of 10mhz Z80 without wait states.  While an
8mhz z80 machine has
limited need when you add buffering and propagation delays you do need
to be under 100ns
for margin.  At 10mhz you have to be under 85ns.

Also back then the largest byte wide was 2116/6116 (2kx8) and they were
fast at maybe 200ns.
The 62256 was later and early parts were barely 150ns.  Now finding
parts that slow is a challenge.
The need for external cache for 386 and 486 based machines drove it way
down such that 15-25ns
static CMOS (mid to late 1990s) became easy to find.

Allison