My Part in the ARM Story

I was privileged to be a part of the original ARM development. Initially I worked with Steve Furber  and Sophie Wilson on the ARM architecture and later ran the silicon implementation team.

Steve and Sophie were devotees of the 8 bit 6502 and knew the architecture and instruction set intimately. The successful BBC machine had been based on this. In 1984 they were greatly surprised by the the poor performance of 16 bit processor offerings and wanted to build something in house based on a paper from Berkeley discussing a research a single chip Reduced Instruction Set Computer. (RISC).

Actually, any processor Acorn built had to be simple. The company traditionally struggled with gate array designs and the BBC chips were on there 14th iteration. Myself, and a number of folks from TI, had been hired to help with this under Mike Hill (our old boss from TI). Our full custom capability was initially just Jamie Urquhart and myself.

I worked closely with Steve and Sophie on defining the ARM. They did the much of architecture. I focused more on overall compromises between complexity and ease of implementation. My contributions were:

1. Simplified clocking scheme. I didn't like the four phase clocking scheme used by Berkeley and, in fact, most 16 and 32 bit MOS processors in those days. This would mean having a separate custom (and probably bipolar) device to manage the clock widths accurately. I also hated the idea of dividing up operations with registers artificially - it seemed most architectures did it.
   
   A simple two phase scheme was what I learned in school! The problem with this scheme is how to arrange the logic so that there was no dead time in the cycle.
   
   The solution for the ARM was to place several CPU operations into each half clock and allow the data to flow asynchronously during each half clock period.
   
   Actually, before starting to work with MOS CPUs I assumed that most architectures worked this way.
   
   
2. Flow through (asynchronous) data path. Per the above the idea was to set up the maximum of un-clocked data flows. So, for example, for register to register operations two registers were read onto busses (statically) and connected to data processing elements (ALU, shifter etc). The second half cycle was used to write the result back to a third register.
   
   By combining static and dynamic logic it was possible to have the advantages of both the dynamic logic and the asynchronous flow through. But this required CMOS logic ...
   
   
3. CMOS Process. It is not generally known, but the ARM was the first 32 bit CMOS processor. We were lucky that in 1983 CMOS was just becoming available for customer designed custom chips. We had the luxury of not having a history in HMOS design and therefore little to loose by making this choice.
   

During the initial design process I considered the ARM to be rather clunky. I was very concerned about excessive power dissipation and even did a calculation to see if the VDD wires would melt during startup. I was also concerned about chip area. The original ARM was 60mm². This was very large for a high volume part in those days.

Hence the extreme focus on simplicity and low power. Of course I was quite wrong. We were all surprised by the power dissipation of the sample parts.

Initial Licensing

I was aware that a new CPU requires a support community for it to prosper. It was obvious to me that Acorn alone could not make the Architecture a success. At the time there was little support for this view in the company. Management mostly took the view that to license the ARM was to give away it's competitive advantage. However, one of the advantages of a chaotic environment is that it is possible to move forward with a non-approved project.

Hence, on the pretext of "let's see what the value of a license would be" I start contacting possible customers. Most did not take this seriously. In those days I had very little gravitas. VLSI Technology (now Philips) was the exception. I was able to negotiate an initial licensing deal capped at $1M for the first 3 years. 

Shortly after this Acorn was bought out, and fortunately the new owners did see the value of the license and went through with the deal. 

Of course hats off the Acorn's culture. I can't think of any other company in the UK that would have gone ahead with this. Although Acorn as a company had many faults, Herman Hauser did create a unique environment that made the ARM possible.

Page created by: rh@ot1.com
Changes last made on: 11/28/06

HomePage