View unanswered posts | View active topics It is currently Sat Feb 17, 2018 7:39 pm

Reply to topic  [ 4 posts ] 
 Motorola's 68020 - faster, bigger, more complex 
Author Message

Joined: Wed Jan 09, 2013 6:54 pm
Posts: 870
Some interesting information about the 68020, from 1984.

First, a blog post with some patent references:
The Motorola MC68020 holds a special place in my heart because it was used in the Amiga 1200, which was the first computer I owned which was a pleasure to use and program.

Next, some stats from a 1988 book about the device, by Werner Hilf and Anton Nausch, also Motorola employees:
    68000 had 68,000 transistors, chip size 6.2 by 7.1mm.
    68020 had 200,000 transistors, chip size 7.1 by 7.1mm.
    2¼ micron HCMOS-III technology, 5ns propagation delay per logic gate.
    Drawings for the chip layout were 150 square meters in total.
    Schematics were 46 square meters in total.
    Test software for the chip had half a million lines of code.
    A "prototype" was built from 300 MSI (medium scale integration) chips on 14 PCBs.

- M68000 Familie Teil 2 Anwendung und Bausteine (translated freely from the German text.)

There's also a detailed 1984 paper The Motorola MC68020 by Doug MacGregor, Dave Mothersole, and Bill Moyer, all of Motorola, which gives slightly different numbers - possibly an earlier revision in a less advanced technology:
The MC68020 is 9.22 mm on a side. lt was designed with an HCMOS technology to minimize power consumption (it consumes 1 to 1.5 watts) and allows for high clock frequencies (16 MHz, for the worst case). lt utilizes about 190,000 transistors, 103,000 of which are actually implemented.

What's an unimplemented transistor? Most likely it means they've counted every transistor site in the ROMs and PLAs, where on average only half would be present, depending on programming.

The primary design goals of the MC68020 were compatibility and performance. The chip's performance was significantly increased through the use of many different techniques-parallelisrl, pipelining, increased clock frequency, increased bus width, special-purpose units, new instructions and addressing modes, and an instruction cache.

Increased clock frequency. One of the simplest ways to increase performance is to increase the clock frequency at which a part executes. The MC68000 and its derivative, the MC68010, have a frequency range of 8 MHz to 12.5 MHz. On the MC68020, the base frequency has been increased to 16 MHz, and it is is not unreasonable to expect 20 MHz or above in the future.

It is now possible to build a machine that is very fast internally, but that machine must still communicate with the external world. Thus it is important to minimize the delay due to executing bus cycles. This is done by reducing the minimum bus cycle time from four clock cycles to three clock cycles, and by reducing the number of bus cycles to be excuted by providing a 32-bit data bus and a256-byte instruction cache. The instruction cache reduces the execution time in two ways. First, it provides a two-clock-cycle access time for an access that hits in the cache, and second, if the access hits in the cache, it allows simultaneous instruction and data accesses to occur. Of these two benefits, simultaneous instruction and data access is more significant since it allows a 100 percent reduction in the time required to access the instruction rather than the 33 percent reduction afforded by going from three clocks to two clocks. Reducing cache access time further (below two clocks) would not be beneficial, since the time required to update the instruction stream pointers would be two clocks as well.

Lots more interesting reading in that paper, and several interesting tables and diagrams too. Notably the 68020 also extends the instruction set: there's an example of an array access, which takes 79 clocks on the 68000, 50 clocks running the same code on a 68020, and just 32 clocks using the new addressing mode.

Mon Nov 20, 2017 7:02 pm

Joined: Tue Dec 11, 2012 8:03 am
Posts: 233
Location: California
I often wonder what kind of design team it takes to make something like this succeed; ie, how the various portions of the job are divvied up, what kind of expertise they have in each portion, how it is coordinated and integrated, how problems found in testing are figured out when it's too big for any one person to have a handle on the whole thing, etc..

_________________ lots of 6502 resources

Mon Nov 20, 2017 8:47 pm
Profile WWW

Joined: Wed Jan 09, 2013 6:54 pm
Posts: 870
Possibly as few as a dozen for the 68020, surprisingly:
MC68020-team.png [ 90.88 KiB | Viewed 409 times ]

- from VLSI CAD Tools and Applications edited by Wolfgang Fichtner, Martin Morf

Call it two or three dozen to include circuit design and layout, especially in those days.

I've seen a few processor design startups, these were much more complex CPUs and indeed SoCs. These efforts grow to 100-250 engineers, but from a very small core of people who really understand what the problem is that is to be solved. You'll need (many) more people if you have an old-school approach, but we managed with small teams. Off the top of my head, let's say about a dozen HDL people, a dozen verification people, a dozen layout people. A dozen systems people (boards), a dozen applications people (logic analysers). A half-dozen toolchain people, a half-dozen algorithms people, a couple of dozen software people, and two or three dozen applications people. A half-dozen people for test design and production coordination. (Very few writing production tests because of automation and use of scan chain techniques.)

Dozen dozen dozen, it's stopped looking like a word.

Mon Nov 20, 2017 9:16 pm

Joined: Wed Jan 09, 2013 6:54 pm
Posts: 870
Adapteva wrote an article on the costs of running a semiconductor startup, and the implications for product pricing and volume: ... tup-model/


Their aim was to find a way to do it more cheaply than the average.

On average the SOC startup team seems to ramp up to approximately 75 engineers quickly and stay at that team size until eventually ramping up towards an exit or going out of business.

We used many of the techniques discussed to reach break–even with less than $2M of total investment.

(Via the discussion here on SiFive's RISCV SoC offering, priced initially at $999)

Wed Feb 07, 2018 6:30 pm
Display posts from previous:  Sort by  
Reply to topic   [ 4 posts ] 

Who is online

Users browsing this forum: No registered users and 3 guests

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB® Forum Software © phpBB Group
Designed by ST Software