31 Jan 2018
There are a number of electronics hobbyists out there who build their own computers - and I don't mean by assembling boards - I mean hard-core design in which they actually design their own central processors (CPUs) and build them from vintage (1970's) TTL logic chips. This started me thinking about the technology factors that kicked off the personal computer revoltion in the mid-1970's
The usual telling of the tale points to intel's introduction of the 4004, 8008, and 8080 microprocessors in 1971, 1972, 1974. From various thing I have read about that time, however, it is not apparent that intel (or many other people) viewed these devices as replacements for minicomputers - but rather as ways of replacing the discrete logic that was ubiquitous in automatic controls in that era. And in some respects, with a $360 initial retail price - for just the CPU chip with no additional support (1974 dollars), it wasn't all that clear that it would even be a very economical replacement for discrete logic in a lot of applications. At $360, it wasn't even all that clear that the 8080 would be a good alternative to some of the 16-bit minicomputers being offered at the time. for example, in mid-1973, Computer Automation was offering an entry-level version of their 'Naked Mini' 16-bit control computer for under $1000 in quantity.
The personal computer revolution, however, did not start in the minicomputer industry, but rather was initiated by a number of hobbyists and hackers. Steve Wozniak and Steve Jobs and the Apple I and II are the most famous, but there were many other individuals and small businesses offering personal computers in the mid 1970's. The seminal milestone was perhaps the Altair 8080 which appeared on the January 1975 cover of Popular Electronics magazine. This computer was based on the newly available intel 8080 single-chip microprocessor (released in April 1974), and was offered both in kit ($439) and assembled ($621) form. The basic kit shipped with 256 bytes of RAM - to do much of anything useful with this computer one needed to buy additional expansion cards with more memory and I/O options.
The Altair 8800 was not, however, the first microcomputer or computer kit. Six months earlier, in July 1974, the Mark-8 mini-computer appeared in the competing hobbyist magazine Radio Electronics as a constructon project (with circuit boards available from the article's author). This project was based on the older 8008 CPU and also came with 256 bytes of RAM - but does not appear to have been designed to be much more than a demonstrator/trainer type of system, lacking add-on cards and expandability.
Although the above machines used the recently available intel microprocessors, by the mid 1970's people had been building small computers for a decade using small scale TTL (Transistor-transistor Logic) integrated circuits. Not only were minicomputer companies doing this, but so were hobbyists. In 1971, the Kenbak-1 first became available. Like contemporary minicomputers, this machine's CPU was constructed using these TTL chips. Performance, however, was significantly lower - largely due to the use of serial shift registers that implemented the machine's 256 bytes of memory. Fast semiconductor RAM (Random Access Memeory) was just becoming available around this time, but was still more expensive than the magnetic core memory in common use in minicomputers and mainframes. While core memory was a mature technology by 1970, it was neither cheap on a per-bit basis, nor was it simple to use - requiring all kinds of precision analog and mixed-signal circuitry to make it work. A semiconductor shift register memory, while relatively cheap (at the time) and simple to use, had the major drawback that it did not provide random access to the stored data. Shift register memories circulate the data on a first-in-first-out basis, so to get to a given byte may require reading every other stored byte first, depending on when the data of interest was was written. making it very slow when used to simulate a random access memory.
What I find interesting about the Kenbak-1 and Computer Automation's early 1970's offerings was that the prices were approaching the range where the products might become interesting to hobbyists - and they did not use single-chip microprocessors. Alternate-history speculation can be fun. What might have happened if the development of the microprocessor had been delayed for a few years? At the time, these iniital development projects were relatively expensive, high risk from both technical and market standpoints, and were only being undertaken by a relatively few pioneering companies. What if no one had bothered?
One question I have is how much it would have cost to make a decent discrete TTL CPU in the 1975 timeframe? While digital integrated circuits were invented in the early 1960's, they were ungodly expensive - and initially used in military and space applications where money was no object. Leading mnicomputer manufacturers like Digital Equipment Corporation continued to use discrete transistor designs through the mid to late 1960's, presumably for cost reasons. While I don't have ready access to manufacturer OEM pricing for semiconductors from that period, I can see what a lot of the hobbyists might have paid through archives of magazines from those periods, with a focus on the mail-order advertising.
The American Radio History website has scans of old hobbyist magazines going back to the 1950's - a major nostalgia trip for electronics buffs. Keep in mind that in the early 1970's, there were no hobbyist computer magazines of significant circulation, so the ones I looked at had more of a Radio/Audio/Widget orientation - with the occasional digital project thrown in. Keep in mind that the products that would have been advertised in these hobbyist magazines would not have been state of the art, but would have represented what a hobbyist could easily obtain at the time . So let's begin a trip down hobbyist nostalgia lane...
January 1970 - Poly Paks was selling 900 series RTL (gate or flip-flop per package) for $1-2 each. It would have taken several hundred of these chips to build even a rudimentary 8-bit CPU. I used to buy stuff from Poly Paks in the 70's - but not digital ICs ! Cost of a minimal useful CPU ? Maybe $500.
January 1971 - The $1 RTL chips have dropped in price to $0.30-$0.50. Also seeing a few '7400' sereis TTL appearing in the $1-$2 range, offering 4-6 gates per package as well as improved performance and ease of use over the RTL devices. Our hypothetical minimal CPU might now contain 150-200 chips at a cost of $300 (retail). I don't see any memory technology for sale, except for some 'surplus' core memory - which as previously pointed out would have been a joy to make work.
January 1972 - Starting to see a decent selection of 7400 TTL, including some more advanced functions like counters, decoders, shift registers and even the 74181 ALU (Arithmetic Logic Unit) chip for $8 each. This is especially significant because four of these devices can be used to implement a minicomputer-performance 16-bit artihmetic unit - the heart of the CPU. Simple gate-type functions are now selling for $0.35-$0.40 each. While the total cost of the CPU might not have dropped much, it is probably now possible to implement a fairly decent 16-bit processor for $200-$250 of ICs. And some other key components are beginning to appear as well - LEDs (~$0.50 each) , 7-segment displays (), a character generator ROM, which makes it easier to implement a video display ($25), and 256 bit RAM memory chips for only $10 each. At that price, 4k of memory chips would cost $1280 - far more than the CPU.
At this point (early 1972), it would certainly be feasible for someone to offer a modest CPU for a few hundred dollars - as we can assume that once they geared up production they could get much better prices than that shown in the back of a hobby magazine. Getting a usefully-sized memory however, is still going to be prohibitively expensive. Let's keep stepping through the years...
January 1973 - Both TTL and memory prices have dropped about 50% - with simple SSI TTL chips going for ~$0.25, 256 bit RAM chips $3. The 1103 1k DRAM chip is now appearing for ~$8. although more complex to use than static RAMs, it lets you put 4k of memory on a single (small) circuit board with ~32 chips at an IC cost of ~$256. We can probably implement a decent 16-bit CPU for ~$150 of silicon. On a side note, I see an early ad for my favorite distributor (Digikey) appear!
January 1974 - Neither TTL ICs nor memory chips have gotten any cheaper in the past year, but there seem to be almost as many ads for digital parts as analog ones. Hobbyists may not be designing CPUs at this point, but it sure looks like they are building other digital projects.
January 1975 - TTL prices have continued to drop to under $0.20 for simple ICs and intel's 1101 256 bit SRAM chip is now under $2 each, but the new 1kbit 2102 SRAM chip is only $5 - this 4X increase in density allows 4Kbytes of SRAM to fit on a single PC board. Another major piece of the puzzle, programmable Read-Only Memories (ROMs) are beginiing to appear. These are important as they allow you to embed permanent 'boot-loader' code into a computer system, which allows a computer to 'boot' from some external mass storage device (disk, cassette, paper tape). If you do not have a boot-loader in ROM, then you have to enter it through toggle switches every time you turn the machine on - a major pain in the ass. Having programmable ROM is important because it lets the-guy-in-the-garage define its contents - the alternative being mask-programmable ROM whcih requires the contents be written by the IC supplier and you have to cough up several thousand dollars of tooling - and pray you don't make any mistakes.
So, by January 1975, it looks like our alternate reality personal computer entepreneur has the following technology items available:
Digital logic sufficient to implement a decent CPU for ~$125-150 of chip cost.
SRAM (intel 2102) at a cost of $40/kbyte.
Character generator ICs for video displays.
LEDs and other display technologies.
The only thing missing in alternate 1975 is the $360 microprocessor - which I suspect would not deter alternate-Woz or his equivalent for very long. The alternate-universe Apple I might have needed 2 PCBs instead of one, but that would be but a speed-bump. From the above look at component availability and pricing (granted 'retail'), my guess is that the introduction of semiconductor memory technology is really what kicked off the personal computer revolution - and not the availability of microprocessors. If microprocessor development hadn't occured for a few more years, I suspect hobbyist-type personal computers would have still appeared in the mid 1970's mainly because enough of the necessary pieces were still there.