The Story of the First Microprocessors | Hacker News Reader

The Story of the First Microprocessors

spectrum.ieee.org

103 pointsjonbaer20 comments

Masatoshi Shima deserves more credit than anybody. He was responsible for most of the logic in the 4004 and the 8080. Should write an article that summarizes what goes on in these oral history discussions:

4004: http://archive.computerhistory.org/resources/access/text/201...

8080: http://archive.computerhistory.org/resources/access/text/201...

z80: http://archive.computerhistory.org/resources/text/Oral_Histo...

Pasted the wrong link for 8080: http://archive.computerhistory.org/resources/access/text/201...

Here is a factoid for the youngins ... the Internet/Arpanet was created BEFORE the first microprocessor! In fact, Intel was originally founded to make RAM ICs. They only later created the first microprocessor (the 4004)!

Back then, it really was a bunch tubes.

Here's a PDF describing the system that Lee Boysel demoed during the TI case:

http://corphist.computerhistory.org/corphist/documents/doc-4...

Boysel's demo is an amazing hack, but in no way is the AL1 chip a microprocessor. The demo seems convincing - he hooked up an AL1 "CPU" with RAM, ROM and I/O and ran programs on it. But I looked in detail and there are some problems.

The AL1 chip is an 8-bit arithmetic/logic chip with some registers. It does no instruction fetching, no instruction decoding, and doesn't have an instruction set. It doesn't implement any control functions. It doesn't perform memory or I/O operations. (It's similar to a 74181 ALU chip with registers.) So how can it operate as a CPU?

The trick in the demo is the ROM has a "ROM memory address registers", which seems like an innocent latch. But this latch is under the control of the ROM, not the "CPU". The ROM and latch form a state machine that is controlling the system, performing RAM and I/O operations and directing the AL1 chip to perform ALU operations. The ROM is not sending instructions to the "CPU"; the ROM is running the show. The ROM doesn't hold AL1 instructions; it holds crazy microcode-like sequences that get the system to stumble through operations. There's no program counter as such, just the ROM jumping from address to address.

In other words, if you think of a microprocessor as ALU + control, the AL1 chip has only the ALU half, not the control half. The ROM-based state machine provides the control half.

I'll also point out that the AL1 die photo [1] from the demo is kind of bogus. Near the bottom is the label "instruction register 23 bits" - the chip has no instruction register and the label is dropped on top of some clock lines. Saying the chip has an instruction register makes it sound much more like a CPU, but it's just fiction.

To conclude, I am extremely impressed that Boysel was able to get the demo working. But don't be tricked into thinking the AL1 is a microprocessor.

[1] http://www.computerhistory.org/revolution/digital-logic/12/2...

Aha! Thank you for the explanation. It works a bit like the Apple II disk controller, then.

Ha! That's not an analogy I would have thought of, but yes, it's the same ROM + latch state machine architecture as the Apple II disk controller. The demo's state machine generated a lot more control signals (24) and controlled the RAM, I/O and the ALU chip rather than a disk drive, but the concept is very similar.

(Background: Wozniak's design for the Apple II disk controller used very few chips. A ROM and latch implemented a simple state machine that did much of the control sequencing. The CPU also did a lot of the low-level control.)

Thanks for the details. Did the TMC1795 have an instruction decoder? I did an google search and could not turn-up any details. From the image it looks like it could be broken-up in two regions, but I'm not sure. Maybe you discovered some details from the patent docs?

Yes, the TMC 1795 had an instruction decoder. Architecturally, it's a very normal 1970s microprocessor. (The 4004, in comparison, is kind of strange.) The one thing that's unusual from our perspective is the subroutine stack is on-chip, with 8 levels.

If you look at the image, the chip can be split into three horizontal slices. At the top is instruction decoding. In the middle is the ALU. At the bottom is the register file. TI originally designed the processor as three chips, and this remained visible in the final chip.

There's very little written about the chip, so I plan to write about its architecture some time. The patents are very detailed (TI is good that way), and explain almost everything if you take the time to go through them.

Here's a block diagram of the chip: https://goo.gl/photos/TRXRc2VKhQNaVxhG9

Thanks again, looks like it was the decoder and register in the low res image of the die that I saw. I eagerly await to read your article detailing the 1795. I suppose the 8 level stack was to allow it to be used without RAM. Also didn't the AGC use an approach a bit similar to what Boysel did with a ROM for many instructions built from smaller instructions? It also reminds me a bit of a TTA.

I think the on-chip stack was a historical artifact. The Datapoint 2200 is the TTL-based computer whose architecture was copied for the 8008 and TMC 1795. The 2200 used serial shift-register memory [1], so writing the PC to a stack in memory would be very slow and inconvenient. Instead it used 8 words of (expensive) RAM as the stack for system calls. The 8008 and TMX 1795 copied the Datapoint 2200 architecture, including the 8-address stack. Intel moved to a "normal" stack in memory with the 8080, which is in many ways a cleaned-up 8008.

Another interesting historical artifact from the Datapoint 2200's shift-register memory and serial processor is that when you're operating on one bit at a time, you want to start with the lowest bit, so you end up with a little-endian processor. For compatibility the 8008 was little-endian, and thus x86, and thus the processor you're probably using now.

[1] http://www.righto.com/2014/12/inside-intel-1405-die-photos-o...

Federico Faggin went to the University of Padova. I guess that's not of interest to most people, but having spent much of the past 15 years there, I always thought it was kind of cool.

The transistor was invented in 1925 by Julius Edgar Lilienfeld (a FET). The first one built was in 1947 (a BJT).

A metal-oxide (MOS) transistor is a FET. Commonly known as MOSFET.

No mention of Mostek, RCA, Rockwell, Motorola, or Signetics.

They do mention MOSTEK in passing, but this article is literally about who created the first single chip microprocessor. Those other companies aren't even in the running.

At least Pico/GI got a mention. I grew up in that area during this period, and my Dad knew a few of the engineers through his work. Also I rode the train to school with a girl around my age who I (much) later discovered was the daughter of the lead designer for their CPU. If only I had known how cool she was ;)

https://en.wikipedia.org/wiki/Microprocessor#Pico.2FGeneral_...

http://www.spingal.plus.com/micro/

Another important thing to note is that the story of the first microprocessors is one that was largely galvanized and thus dominated by US military investment in the 60s. This bankrolled research in those days, getting ICs out of small-/medium-scale integration (< 500 transistors) into large-scale and onward - at which point Intel, Mostek and others showed up on the scene (circa 1970) to target the commercial market.

Or Zilog.

Zilog the company wasn't founded until 1974, years after the first microprocessors were made.

Reply on news.ycombinator.com