I've just finished reading the documents from those links. Very interesting, although not very entertaining reading - these are raw transcripts from the discussion, so there are a lot of colloquialisms, umms and emms, and also several errors and typos introduced in the transcription process. However, the facts are interesting, and they also clear up the timeline a bit.

The 8048 was apparently not the first single-chip microcontroller - Intel had its 4-bit microprocessor line by that time (4004 and 4040) and maybe also the first 8-bit microprocessor (8008), but the microcontroller was defined as a competitor for the TI's TMS1000 and also Mostek's 3870 (a single-chip version of Fairchild's F8) was already being advertised (although maybe not quite ready) at that time. But what the novelty was, was the introduction of EPROM as program memory - remember that it was Intel who introduced the first EPROMs, right around this time. This made it easy for the developers to prototype complete products rapidly - the competitors were all ROM based, so the final chip took ten to several tens of weeks of semiconductor fab turnaround to make, not to mention the costs of a programming error. They were also targeting for a single-supply-voltage technology, which was not that common in those times. The technology to achieve that in those times was PMOS.

So, the real target was to make the 8748 rather than the 8048 (which was seen as secondary). The ROM-less 8035-s were 8748-s which EPROMs failed during the production testing. There were also the '41/'42 developed simultaneously, which had a slave parallel interface - this was sort of a political thing within Intel, so that the groups developing the 8080/8085/8086 would see the microcontrollers as an intelligent peripheral to the microprocessors, i.e. as a complementing line, rather than competing.

Many of the engineers came from the MCS-40 background and the '48 was seen as a "low end" product basically replacing the "low end" MCS-40 (the 8- and 16-bit processors being the "high end"), so there was pressure on cost (targeted at $10 initially, and to third or quarter of that in production) hence the silicon area. However, the designers really wanted to make it 8-bit, as they felt 8-bits are more sexy. That meant to make certain tradeoffs in the design, to spare as much silicon area as possible, to make only those instructions that were really necessary - so for example there are no subtract and compare instructions in the '48, and that is also the reason for the relatively small 2k addressing space (which has its legacy in the '51's AJMP/ACALL) and no relative jumps. The technology at that time was at around 5um.

There is an interesting story about the instruction set mentioned: "I also want to point out Dave's two famous - or his favorite instructions were SEX and SIN, which stood for set external mode and set internal mode. And then when the people in systems took over they removed [renamed] SEX and SIN." (I could not determine which of the final instruction set these were...?).

There were no computer tools (CAD) except a circuit simulator, and even that run on a mainframe computer shared with the financial dept., so the engineers were not able to work by the end of month... Layout was drawn by hand, with no design rule check tools, and then hand cut to rubylith (a two-layer plastic sheet, one opaque (red) to be cut and peeled off the other transparent), as a base for the mask (after photographic downscaling). The developer recalled he wanted to get the best "rubylith peeler", as bad quality of "peeling" resulted in various hard to find bugs.

Naturally, there were errors in the first silicon. But before that, they have built a breadboard version of the '48 from TTL chips - five foot tall times three foot wide. So when they got the first chip, they cut the traces inside the chip and then connected various parts of the chip to the breadboard, try to run that "hybrid", and so they determined, which part of the chip worked and which did not.

The second silicon did already work OK, which was quite a feat those times. That was by the end of 1976, and in 1977 the 8748/8048 was actually being sold. There were initial troubles with technology - they did not run at the specified speed, for example, and the production yield "was atrocious". But they eventually get it right soon and the chip performed quite well in the few next years.

There is one more nice story there, when the designers got to build a demonstration application to show to the top management. They chose to build a blackjack machine, attached to a dumb terminal. But what was interesting was the power source: "So we took like a lemon or an orange, I don't remember, with two dissimilar metals, and we stuck it in there."

---
Very soon it was clear that the tradeoffs made with the '48 design - mainly the limited address space - is not going to meet the future demands, so they started to summarise the requirements for the new design as soon as in 1977, the very year in which '48 was introduced. Besides "correcting" the defficiencies of '48, also the idea of managing the peripherals through SFRs (in '48, they had specific opcodes, which required slightly different instruction set for every model), and the boolean capabilities were added. And, of course, the succession to '48 was stressed, to keep the customers already using that one. By the beginning of 1978, a specification was born, i.e. that year can be seen as the moment of "conception" of 8051. Even if there already were CAD tools available, it took until the end of 1979, however, to get it to the first silicon (partially maybe because meantime the microcontroller group was moved from California to Arizona) - and it took 3 more revisions to get it right. The interesting thing is, that the stepping before the first good one was already used for demonstrations, although the conditional jumps did not work on that - the boolean capabilities did, and a real appnote (AP-70, "Using the Intel MCS -51. Boolean Processing. Capabilities", by John Wharton) was demonstrated on it. According to the article, first real working silicon came in 1981 (so the "official" date of 1.1.1980 is maybe a "midpoint" between the first specification and the real availability of the chips).

Another interesting fact is, that the first silicons were NOT the production 8051; rather, it was a bondout chip - this allowed not only to debug the internals of it faster, but also to have an ICE ready at the same time when the real chips came out. That was one more step on Intel's strategy to make it easy for the users to prototype - besides offering the EPROM chip.

The rest is, as they say, history.

JW