News & Analysis
Why the ARM architecture is shaped the way it is
Peter Clarke
11/26/2012 9:25 AM EST
It is arguable that ARM and Intel, the two companies locked in head-to-head processor competition, represent two different poles and philosophies.
ARM is modestly British while Intel is brashly American. Intel's chips have been dominant in personal computers for decades while ARM's architecture has been market leader in cellphones and smartphones. ARM is skilled at low power and power efficiency and is trying to move up to higher performance. Intel has always pursued performance but is now working to offer low power versions of its microprocessors. Intel is an integrated device maker. ARM does not even sell chips – but licenses out its processor architecture and circuit designs to semiconductor partners.
To understand why the ARM architecture and culture is shaped the way it is and is different to processor trailblazer Intel, let's go back to a time before the formation of ARM; to Cambridge, England in the mid-to-late 1970s, in the early days of EE Times.
Sophie Wilson is now engineering director and distinguished engineer at Broadcom but back in the 1970s Roger Wilson was a mathematics undergraduate with a strong interest in computing.
Wilson had joined the university's computer and microprocessor societies, student-run interest groups that held regular meetings and, in the case of the microprocessor society, produced a magazine.
"One of the articles I contributed was on low power but it was not my only area of interest," Wilson told EE Times.
However, Hermann Hauser, a Cambridge physics alumni, who was thinking what to do next in his career had the idea of creating an electronic pocket diary. Obviously, that would need to be battery operated and low power. He therefore asked around trying to find out who at the university might be able to design such a thing for him and Wilson remembers being approached after a university lecture by contacts of Hauser's.
Wilson had memorized the RCA CMOS logic handbook of 1972 at a time when most ICs were not made using CMOS. Initially CMOS logic was slower than TTL. However, because the logic thresholds of CMOS were proportional to the power supply voltage, CMOS devices were well-adapted to battery-operated systems with simple power supplies, in which the voltage might vary. "CMOS was graduating from 15-volts to 5-volts," recalls Wilson, which provided further power saving opportunities.
The result was that when Hauser asked whether Wilson could design such a low-power system the second-year undergraduate student simply said "yes!"
Click on image to enlarge.
Sophie Wilson
Wilson started doing designs without any form of contract, just because that is what members of the computer and microprocessor societies liked to do. Each had their own favorite microprocessors and at this stage in the 8-bit era there were choices. Wilson had experience of designing around the 6502, including an early form of embedded system; an electronically controlled cow feeder for a firm in Harrogate.
The 6502 was designed by Chuck Peddle and Bill Mensch for MOS Technology Inc. in 1975 and when it was introduced was considerably lower cost than equivalents from Intel and Motorola. Mensch went on to form Western Design Center Inc. (Mesa, Arizona) which made the 65C02, the CMOS version of the processor.
At about the same time another Cambridge undergraduate was working for Science of Cambridge Ltd. on the Microcomputer Kit 14 (MK14) home-build computer, first introduced in 1977.
Science of Cambridge was one of the companies operated by Clive Sinclair (now Sir Clive Sinclair) whose Sinclair Radionics company had enjoyed mixed success in the UK supplying hi-fi audio electronics, and some of the earliest pocket calculators. The MK14 was based on a National Semiconductor 8-bit processor and the undergraduate was Steve Furber, now Professor Steve Furber of Manchester University.
Wilson remembers that it was while discussing plans for the electronic pocket diary Hauser pointed to other plans at the back of the sheaf of papers. These were Wilson's own plans for a personal computer. "At the microprocessor society we all had our own designs," Wilson recalls.
Wilson remembers that at this time the MK14 was starting to get press coverage and it seemed like personal computers might be about to take off. Hauser asked if Wilson's personal computer design would work. Wilson simply said "yes." That was that design that went on to form the Acorn System 1 personal computer – based on the 6502.
Click on image to enlarge.
From tiny Acorns: Front board of an Acorn System 1
Next: Acorn roots
|
|
|
|
|
Navigate to related information
EREBUS
11/26/2012 3:35 PM EST
I give ARM good marks for seeing a development path beyond the Intel line. Intel followed a logical path exploiting their processor line into the home PC market.
ARM saw the long term potential of building a path towards customizable computer processor components to enable smaller runs of targeted processors.
I am impressed with the quality and versatility of the ARM processor line. They may now face more competition from Intel as the PC market begins to phase down. The capability to make building block components has been around for a while, so it will be interesting to see if Intel can come up with a competitive alternative.
Sign in to Reply
Duane Benson
11/27/2012 12:40 AM EST
Nice article. I was around when much of this was happening and had heard a bit about Acorn computers but was mostly unaware of how it all fit together.
I especially like this statement: 'Wilson concludes: "Hermann Hauser says he gave us the things Intel could never give us, no resources, no time and no money."'
How many great innovations have happened because someone had a job to do and not enough time, resources or money to do it?
Sign in to Reply
chipmonk
11/27/2012 11:29 AM EST
From tiny acorns do giant oaks ( ARM ) grow ! Ironic that de-industrialized and uncompetitive England provided a more fertile soil for RISC designs to grow than out here in the desert in the shadow of giant Fabs that are still churning out CISC processors with a billion transistors. But for how long ?
Sign in to Reply
John_Galt
11/27/2012 3:48 PM EST
Thank goodness for those 10 Billion transistor giants. Want to try running SolidWorks or AnSYS' Maxwell on even the most powerful ARM8 (if they were even available)? Comparing ARM and Intel architectures is like comparing helicopters with automobiles - they both have areas where they are a great fit - but neither address the full universe of applications.
Sign in to Reply
przemek
4/22/2013 5:19 PM EDT
Shrug---it's more complicated than that, my dear Atlas.
If you squint at published benchmarks like Specmark just right, the more complex ARM models gets comparable or better performance per GHz. The real reason why x86 architecture runs your favourite applications is, as you pointed out, because they are not available for ARM, because Wintel.
This is changing slowly: there are reports that PC sales are crashing, and the Wintel shiny front wall starts showing cracks. Will SolidWorks be available and usable on Android any time soon? Probably not, but the reason is not 'more transistors' or better architecture on x86.
Sign in to Reply
jackOfManyTrades
11/28/2012 4:01 AM EST
De-industrialised? Uncompetaive? Take a look at this PWC report:
http://www.pwc.co.uk/assets/pdf/ukmanufacturing-300309.pdf
"There is a widespread assumption that the final demise of manufacturing in the UK is only
a matter of time. But this is simply not so. The facts tell a different story:
• Output of British manufacturing reached an all-time high in 2007, even adjusted
for inflation
• The UK is the world’s 6th largest manufacturer with strong positions in certain key
industries, e.g. a 15% global market share in Aerospace
• UK Manufacturing achieved a 50% increase in labour productivity from 1997-2007"
Sign in to Reply
jackOfManyTrades
11/28/2012 4:02 AM EST
and my country is not called "England" anymore than the US is called "Texas".
Sign in to Reply
John_Galt
11/28/2012 10:33 AM EST
@Jack...go easy on our friend chipmonk. For the past 50 years public education here in the States has atrophied the curricula so badly that many "students" couldn't find England on a map. I work with younger BS/MSEEs who cannot even compose a coherent paragraph...they're not stupid, in fact many are quite bright, it's just that the system has failed them.
Sign in to Reply
help.fulguy
11/27/2012 11:50 AM EST
Nice Ad. How much ARM paid for this crap hmm...Ad.
Sign in to Reply
John_Galt
11/27/2012 3:35 PM EST
They're all good. In today's world of SoC the architecture of the CPU usually, but not always, takes second or third place when comparing features like the peripheral set, Pd and package alternatives. Afterall, all Boole, et al left us with is AND and NOT...everything else is but a variation on a theme.
Sign in to Reply
rogkru
11/27/2012 4:13 PM EST
Was the ferranti ULA a sea-of-gates with a metal top layer - or was it actually an eraly FPGA ?
Google found this... chances are google run on thousands of Intel processors.... ironic.
http://tinyurl.com/ferranti-ULA-problems-delay
Sign in to Reply
Mike Santarini
11/27/2012 5:22 PM EST
Great article, Peter.
Sign in to Reply
TanjB
11/27/2012 5:42 PM EST
ARM in early 90s was not so much in advance of other chips, especially MIPS, and MIPS sold a lot more. Hauser forgot one other essential thing he gave ARM: no customers. So when the mobile industry was ready to build SOCs, there was ARM with a suitable chip, fabless portable design, and no other business model to distract them. It was their one lifeline, and it proved to be a bonanza. The dominant competitor capable of providing a choice at the time, MIPS, was very successful elsewhere and by the time phones were big enough business for them to become concerned, ARM had a lock.
These days, ARM is part of SOCs with billions of transistors just like x86. The difference is in balance: in the ARM ecosystem the die is shared with IP like GPUs, modems, and half a dozen other blocks which have formed a healthy ecosystem. x86 is elbowing in: will it have more success now than MIPS 15 years ago?
Sign in to Reply
bordersboy
12/4/2012 11:48 AM EST
The core of ARM was that it had to be cheaper than an off the shelf processor or else Acorn went bust. Thats why when Robin Saxby took over he needed to find a customer for the core as Acorn were not buying enough to keep going. Nokia told people to licence this as they were not paying for someone elses IP as it would mean GSM handset were going to be too expensive. TI licenced it and there history was written
Sign in to Reply
rick.merritt
11/27/2012 6:31 PM EST
Nice telling of a piece of microprocessor history.
The lesson, Just say, "Yes!"
Sign in to Reply
Kevin N
11/27/2012 10:43 PM EST
The story isn't too clear on this, but Roger Wilson and Sophie Wilson are the same person.
The 65C02! What a nice processor.
Sign in to Reply
dot.my
11/28/2012 6:59 PM EST
Peter, can you make it clear... Roger vs Sophie ?
Sign in to Reply
rick.merritt
11/28/2012 7:09 PM EST
I celebrate some diversity in our industry!
Sign in to Reply
C VanDorne
2/12/2013 11:43 AM EST
What an odd reply!
I think the point is that this is story about people, and at some point in the timeline the main character in the story had a sex-change, which is a remarkable thing, making it a glaring omission from the story.
Sign in to Reply
dot.my
11/28/2012 7:21 PM EST
wiki: Sophie Wilson (born 'Roger Wilson' in Leeds, England, in 1957) is a British computer scientist.
Sign in to Reply
agk
11/28/2012 1:26 AM EST
The article is interesting to read. It is electronics history. While reading visualize the same and many interesting thoughts comes to my mind. Here the heroes live a student cum professional life.
Sign in to Reply
Greg.Overkamp
12/27/2012 12:42 PM EST
agk I have read your comments numerous times and they are an uninspired wasted of time and space for the informed. Please go away.
Sign in to Reply
rogkru
11/28/2012 4:16 PM EST
dramatisation at 0:40:00 point in this...
http://www.youtube.com/watch?v=sIcAyFVK0gE
Sign in to Reply
PR14
12/1/2012 5:27 AM EST
The first project I worked on used the BBC microcomputer to do 3D spherical trigonometry using radio signals in real time to compute location - sort of a poor man's GPS.
What was astonishing was how fast floating point code written in Basic would run on this 8 bit 6502. I guess this was due to hand-crafted C or assembler floating point routines embedded in ROM but I still don't really understand how they made interpreted code run so quickly.
Sign in to Reply
znajit
12/1/2012 1:28 PM EST
Yes I recall my days at ACORN and interaction with Herman, Roger and Steve.
I even recall Herman & my discussion about the daily calendar product that was to have 16 charter LCD displays from Hitachi and use there RT chip who we were connected with very closely for memory parts for our BBC and Electron
What a group and what an innovation company from the day one joined…looking back it was love of my life as I have not come across another company like it for pushing the envelope on daily bases. Gopro may today’s such leader
At ACORN, each one of us was doing 5 projects at any one time from network to making mouse using CCD sensors. I even tacked a research storage device where there head moved spirally to the center with removable 2in cartridge. This is before the 3.5in Discs became the norm
Oh I still have an Electron and it still boots-up and runs Basic, 6502 Assemble like no other product today to hack code in 5min and run 2MHz assembly in Basic. What a concept
Only good memories of my carrier and working with Herman, Roger Wilson (Sophie Wilson) and off course Chris Curry who we started a company called GSI out of his 40 bedroom Manson in Croxton Village together with Ram Banerji
ZahidOh I still have an Electron and it still boots-up and runs Basic+6502 Assemble like no other product today
znajam@msn.com
Sign in to Reply
BrainiacV
12/3/2012 4:08 PM EST
Back in the days I worked at an early computer store, we ran a set of BASIC benchmarks originally published in BYTE magazine on the other computers in our store after we found our numbers agreed with BYTE's list of computers they ran on.
I further wrote a program to sort and rank the results (this being the days before VisiCalc) and one item I tracked (but did not sort on) was the processor the syetem used.
Were were looking to see who wrote the better BASIC interpreter, but were surprised when we saw processor stratification instead.
The 1 MHz 6502s were faster than the 4 MHz Z-80s, followed by the 4 MHz 8080s.
Looking into it, aside from the pipelined instructions, loads took 3 cycles on a 6502 while the Z-80 took 6, and the condition code was set automatically by the load instead of requiring another instruction like the Intels, I felt it was the reduced number of registers (6502 A, X & Y) (Z-80 A, BC, DE, HL + mirror set) helped as well. From my experience programming S/370 Assembler, I had seen too many cases of registers being swapped around, just to make use of register sets already being used, or for special purpose use, which in the end did not do much but chew up clock cycles. Given the reduced number of registers on the 6502, you made use of them for good purpose, you didn't calculate a value and let it hang in a register for K's of execution later until you whipped it out of seemingly nowhere (making debugging harder as well).
That made the 6502 far more efficient than the Intel processors, even though they ran at 4x the clock rate.
Sign in to Reply
Greg.Overkamp
12/27/2012 12:48 PM EST
Nobody gives rat's ass what you used registers for on any CPU. ARMH is successful because customers can design targeted SOCs in 1/4 the time it take Intel to provide a reference design for what they define as the next mobile CPU one year too late.
Sign in to Reply
C VanDorne
2/11/2013 3:17 PM EST
No more caffine for you, Overcamp!
Sign in to Reply
charlie babcock
12/5/2012 3:14 PM EST
Extremely well done story, Peter. I enjoyed it. "It took four clock ticks to run a (68000) memory cycle." That told me something I didn't know about the evolution and inherent advantage of the ARM architecture, which I assume took one tick. Charlie Babcock, InformationWeek
Sign in to Reply
richyoung
12/7/2012 1:14 PM EST
Most processors at the time of the 68000 were CISC or multiple clocks per instruction. It wasn't until later when power and size for embedded applications were the drivers that brought RISC (ARM) type processor to the forefront.
Sign in to Reply
johnswolter
12/6/2012 1:54 PM EST
About the writing...There appears to be a paragraph or two missing at the beginning of the article that sets the scene. The first sentence starts with "Most computers at that time". I guessed the setting was the late 1980s.
I was trying to understand who the players were. Mentioned were " Wilson, Furber and the rest of the team" but no context.
I noticed an inaccuracy about the IBM PC processor. The first PC used the 8088, the 8-bit data-bus version of the 8086.
Wait...I just noticed at the bottom of the article a page button where it says page 3 of 3. I arrived from a LinkedIn discussion group. I don't see paging buttons at the top of the article.
Sign in to Reply
peter.clarke
12/7/2012 7:49 AM EST
@Johnswolter
Glad you found the other pages.
Thanks for pulling me up on the 80286 error. I will correct.
Sign in to Reply
ddaly
12/7/2012 9:03 AM EST
This is more a *very* brief history of the people responsible for the design of the ARM architecture than it is a history of the conception of the design itself. I walk away from reading this article wanting to know way more about the design philosophy and design choices that were made regarding how the architecture came together. To me, the instruction set and the programmer's model, and the thought process going into their design, constitute much of what I would consider the "shaping" of the architecture. That merited one short paragraph.
That the ARM architecture had to be simple, compact, fast, and have low power consumption is a little obvious. That the framers of the ARM architecture have their roots in the 6502 is little more than interesting trivia.
Sign in to Reply
richyoung
12/7/2012 1:08 PM EST
Modestly British?
Brashly American?
Wasn't the future queen topless on the front of the newspaper a few weeks back?
Sign in to Reply
peter.clarke
12/7/2012 2:55 PM EST
@richyoung
I understand the Duchess of Cambridge was caught topless by a long lens photographer, but I also understand it was without her knowledge and that it was published against her wishes.
Sign in to Reply