The 3101 SRAM was Intel's first product. At $99.50 for 64 bits, it had enough memory to store the characters
expensiv. (Sorry, the final
e costs extra.) Is there a record of any product using it?
It might be important to know that the 3101 was neither a genuine Intel development, nor intended as a RAM - at least not in a way we see RAM today. After all, what use could there be in 1970 for a RAM 30 times faster than average core but quite small, just a few words ... hmm ... what data store can be small but should be fast?
The 3101 is Intels Version of TI's 7489 (*1,*2), which in turn was developed as 'scratchpad RAM', an application today more commonly categorized as 'registers'. The 7489 was produced in several technologies (IIRC standard TTL, S and LS types) and many manufacturers (Example Signetics, Fairchild).
What Ken Shirriff doesn't tell in the cited entry is that the Datapoint 2200 not only used 1405 serial RAM but as well four 7489 as register file (*3).
Similar did a large number of mini computers or computer alike machinery wherever a register file/scratchpad was needed. It was essentially everywhere.
It was kind of a sweet spot product for (new) chip manufacturers. Good demand, comparably high prices due being a high integrated one, while at the same time being simple to design and build due its regular structure. No wonder Intel added them early on.
*1 - IIRC first introduced as SN5489 in ca. 1967
*2 - It's often forgotten, but chip numbers are in the first place company specific order numbers. As such they are only valid within that companies order system, different companies use different numbering. That's why there are countless lists of interchangeable types. Using or not using a number from a competitor is a marketing decision - and even then usually only more prominent sections are 'cloned'.
*3 - Yes, four, as the Datapoint 2200 Version 2 had two register sets. The 8008 implemented only one set (like the 2200V1), leaving room for Fagin to 'reinvent' the second set for the Z80.
How would the price and speed have compared with that of e.g. eight 74373 and one 74138 (which could easily be used to build an RAM)? Or if the 74373 was not yet available, eight 74259 addressable latches and eight 74251 multiplexers? Apr 29, 2021 at 19:49
1@supercat it would be with either more than fast enough, as the show stopper was cora cycle time. There aren't many register access cycles needed for any operation before memory timing hits again. More important, building it from discrete latches would need more space and power, neither is a good idea when building machines.– RaffzahnApr 29, 2021 at 20:10
1Was the savings sufficient to justify the price tag? If the 74373 and 74138 were available for e.g. $5 each and the board area and extra cooling would together cost $25, then spending $99.50 to squish the memory into one chip wouldn't be worthwhile, but spending $40 to do so would be a win even if the discrete chips cost $2. Apr 29, 2021 at 21:43
I wonder if the chip was planned to sell for much less than $99, but the initial price was chosen with the intention that people who would be expecting to make many units with it could get started with their design testing before production was in full swing? Apr 30, 2021 at 20:33
1Regarding *3, The Datapoint 2200 Version I had a single register set; this is the version that the Intel 8008 copied. The later Datapoint 2200 Version II had two register sets: Alpha and Beta. Jan 3 at 0:46
The 3101 was not a product that came out of thin air. According to Intel's own sources, Honeywell Inc. had anounced they would buy 64-bit memory chips from any vendor that could supply them (so even from a then unknown newcomer like Intel). (The fact that, eventually, Honeywell did not buy from Intel, is a different story)
So, this product already had a potential customer and a strong market demand before it was even designed.
In addition, 3101 shouldn't be considered main RAM that had to be available in huge quantities. It should rather be understood the same way as fast static RAM chips that were used as cache memory in the first '386 machines - scratchpad memory of a very small size but with very fast access to off-load slower memory (like core memory).
Ais at 65; it leaves room for a space character at 64 in that subset.)