Will Self-timed Asynchronous Logic Rescue CPU Design? - Embedded.com

Will Self-timed Asynchronous Logic Rescue CPU Design?

Like the proverbial reaction of a bull to a red flag, the mere mention of asynchronous logic design prods engineers into heated debate. It will be interesting to see how the computer industry reacts to the revelation this month at the Hot Chips Symposium at Stanford University by Fulcrum Microsystems, (Calabasas, CA) of a MIPS-based self-clocking asynchronous design. However much debate the announcement provokes, will the topic and the issues it raises quickly disappear, as has happened several times in the past? Or will it actually trigger some industry-wide soul-searching about new design methodologies?

Based on the indifference, boredom, and even outright hostility to this idea in the past, my hunch is that the idea will just disappear. But I hope that this methodology, or something like it, finally moves into the mainstream of computer design. Something is needed to pull in the reins on the synchronous horse the industry is riding. Otherwise, we will all ride it right into a wall of inoperability caused by outrageously high power consumption, power dissipation, and global clocking hang-ups as ICs get bigger, denser, and faster.

Illustrative of the lack of public attention given the subject, the presentation by Fulcrum founder and chief technical officer Andrew Lines was the only one of 25 papers at the symposium that addressed the topic. I wonder if the fact that Fulcrum has built a crossbar switch at the heart of its asynchronous design capable of a 260-gigabit-per-second bandwidth in a standard, and relatively conservative, 0.18 micron (180 nanometer) CMOS process sparks any interest.

Unlike the familiar design methodologies used now, self-clocking, asynchronous circuits remove the need for a global synchronizing clock. Instead, the process of computation is controlled though local clocks and local handshaking and handoff between adjacent units. What this means for high performance and power-efficient design is that such local control permits resources to be used only when they are necessary, similar to data flow architecture in the highly parallel designs used in embedded network processors. Although asynchronous designs usually require more transitions on a computational path than synchronously designed CPUs, the transitions usually only occur in areas involved in the current computational task.

Intuitively, the methodology offers a lot of benefits to designers of the new generation of ubiquitously connected embedded devices and small footprint iappliances as well as many embedded networking designs. Because self-clocking allows more fine grained control over the logic, only those portions of the chip that are actively involved in a particular operation are on at any one time. This means that the majority of resources of a circuit can be focused on that chore without driving up power dissipation and consumption, the limiting factors in almost any limited resource design.

With regards shifting to asynchronous design, the electronics and computer industry reminds me of the automobile industry as cars got bigger and faster, but also consumed more fuel and generated more pollution. Then the world started changing. Because of a constellation of factors, the public wanted smaller, more fuel efficient vehicles. While auto makers in the U.S. did finally get the message, they are still dragging their feet, preferring to come up with modifications and enhancements of the existing internal combustion engine.

Just as customers 30 years ago stopped buying big American cars, so have consumers stopped buying 2- and 3-GHz desktop systems. Computer and microprocessor vendors, like their auto making predecessors, are confronting a new world, and, like the husband who cannot understand why his wife has just left him, asks: “What did I do? What did I say?”

For auto makers, 30 years later, it is only now sinking in what they have to do: they are shifting away from gas-only engines, and committing to designs based on hydrogen, battery, power cell and hybrid technologies. Can we afford to wait another 30 years before the computer industry finally bites the bullet and tries something else — anything else — to power our smaller and faster, but more efficient, netcentric information engines?

Asynchronous circuit designs were common in the early days of circuit design and can still be found here and there, not as an all-encompassing design methodology, but as a problem solver, after every synchronous design trick has run out of gas.

By and large, most circuit designs are built with synchronous logic, small blocks of combinatorial logic separated by synchronously clocked registers. The biggest advantage, at SSI, MSI and LSI levels, at least, was that synchronous logic made it easy to determine the maximum operating frequency of a design by finding and calculating the longest delay path between registers in a circuit. AT VLSI and now at the ultra-scale levels with millions of gates it is becoming extraordinarily difficult, requiring much manpower, and computer time, to find and predict the critical path delays.

But that is not the only problem. For one thing there are the large clock current surges necessary, which tax a circuit's power distribution nets, as well as the thermal stability of the circuit. There is also the growing inability to control noise and metal integrity. And in some of today's system on chip designs with not only millions of gates, but millions of heterogeneous gate structures — DRAM, SRAM, ROM, flash, digital logic, programmable logic and analog circuitry — the job of maintaining the global clocking across the area of a chip is not easy.

To be sure, process engineers and logic designers have been quite innovative in finding quick fixes when their synchronous engines show signs of sputtering and dying. They remind me of when my cousin Jerry and I would race around the farm country in rural Oklahoma in his carefully built hot rod. Invariably it would at some point give out and die, whereupon Jerry would jump out, look at the engine and say, with just a little edge of panic in his voice: “It's OK. Don't worry. I got it. I got it.” He would find a clever fix and we would be on our way again, but we were unlikely to get anywhere on schedule.

As we push process technology further down though the micrometer to the nanometer level, things only get worse, with shot noise, charge sharing, thermal effects, supply voltage noise and process variations all making calculations of delay that more uncertain and difficult. Although it is being resisted fiercely as an all-encompassing design methodology, major semiconductor firms such as Intel, TI, IBM, and LSI Logic are selectively making use of asynchronous self-clocking mechanisms in their designs.

Start ups and research efforts at universities continue to work on full designs and the supporting infrastructure to make asynchronous logic more acceptable to mainstream computer design. Several years ago, a research team at the University of Manchester in the UK was able to build a fully asynchronous ARM processor with the performance equivalent of a traditionally designed CPU, and at much lower power.

Even those early designs made it clear, at least to me, that asynchronously designed circuits would be of enormous benefit in what is now emerging as the mainstream of computing in small footprint iappliance design, and many areas of embedded networking, where power issues and reliability are just as important, if not more so, that pure speed.

If the benefits are so clear, why does this “not so new” asynchronous methodology, which was covered extensively in Carver Mead's seminal Introduction to VLSI Systems (Addison-Wesley, 1980), engender so much emotion when I bring up the topic in conversations with engineers?

I don't think it is a technical issue, but an infrastructure support problem. It's the chicken-or-egg question all over again: we cannot easily design asynchronous systems because appropriate tools aren't available. And there are no tools, the EDA houses say, because there is no demand for them.

Break outs from this destructive closed loop are occurring, as basic research continues on tool methodologies at schools such as the University of Manchester, the University of Washington, California Institute of Technology and the University of Utah.

Fulcrum itself is trying to break out of the loop by coming up with a family of MIPS-based SOC cores, which while operating asynchronously internally, will allow developers to design the cores into their synchronously-designed SOCs.

What also makes me optimistic is that EDA tool companies seem to be abandoning their “better the evil you know rather than the evil you don't” attitude. Mentor Graphics over the last few years has nosed around the edges of the problem in a number of internal studies. And Cadence Design Systems is an investor in Fulcrum and a partner with the company in the development of the EDA tools Fulcrum is using.

What do you think? Give me your feedback? Let's get a dialog going on this important issue. I'd like to hear from you.

Bernard Cole is the managing editor for embedded design and net-centric computing at EE Times and the editor of iApplianceweb. He welcomes contact. You can reach him at or 928-525-9087.

Reader Feedback

Anything that reduces the amount of electricity required to operate a computing machine of any kind should be utilized ASAP. How many rolling blackouts must sweep over LA before energy-conservation becomes issue #1? I am a layman RE: design & engineering of CPU's. I don't need to be an expert to see the common-sense that is absolutely evident in “asynchronous” architecture. Thank you for an informative article.

John L. Jordan

Dear Mr. Cole,

Your observations are correct:

  • Synchronous design is a nonsense, and whoever struggled with equation P=V^2*C*f on chip level knows it.
  • Industry will fight better methodology with vengeance.

For some reason, industry has rare talent to embrace the worst solution in sight, make a standard out of it, and support with all brain substituting tools. History repeats itself over and over. One could recall victories of VHS over Beta, or X86 over 68K and so on and so forth.

The only way to alter this regrettable state of affairs is to acquire, or be coerced to acquire great respect for energy. Should salaries be paid in Joules, and should wasting of the energy be punishable by manual cranking of some implement, then the energy per task would regain visibility. In other words, severe energy shortage would do miracle on CPU design, transportation, and almost any other aspect of so called civilized life.

Best Regards,

Tadeusz J. Maydell
Sr. Staff Hardware Engineer
Lockheed Martin

I enjoyed your article “Will Self-Timed Asynchronous Logic Rescue CPU Design?”. Of course I am somewhat prejudiced since Theseus is one of the small companies trying to commercialize self-timed asynchronous technology.

With many of the benefits of clockless technology demonstrated in silicon by the large companies such as Intel, Sun, Philips, Sharp etc., through small companies such as Fulcrum and Theseus, to the many university efforts, you might expect this technology to be gathering momentum rapidly as we push into deep sub-micron processes and systems-on-a-chip.

This is clearly not happening as fast as we would like and I think you summarized the issue very well in that it is an infrastructure support problem. Theseus has been trying to get around this by developing a design methodology based on the use of commercial EDA tools. Currently we use VHDL design capture, Synopsys Design Compiler for synthesis and optimization, and Cadence Ensemble for place and route. This gives us a robust design methodology, but we cannot use many of the Design Compiler optimization rules because they introduce delay sensitivities into the resulting circuits. This then leads to your chicken-or-egg problem with the EDA companies, and so far we have not been able to get much help in modifying the tools to support our design approach.

I would be interested to know what feedback you received on your hypothesis. Do people in general agree with your assessment, or are there other reasons for the reluctance of designers to use a self-timed design style?

David Lamb

One of the areas we (DARPA) have been considering to support the DoD needs in advanced ICs is asynchronous logic. It is my belief that while Intel et. al. may be able to continue to use existing chip design technologies because of the large amount of design resources available to them, the DoD community will not be so lucky.

Given limited resources and complex problems (a radar DSP task has got to be a lot harder problem than a cell phone DSP!), if techniques that allow much more efficient and shorter design cycles are not developed, the DoD technology base will stall at essentially <2005 level. The design problem is critical, but it sure would be nice to have lower power too. And, maybe if we did not have to contend with a giant EMI spike from the clock, maybe we could make more sensitive RF SoCs to detect the bad guys even more effectively.

As in your article, sounds pretty obvious that the scope and limitations of such a technology should be looked at. However, also as you suggest there is the chicken and egg problem. But, we have three of them. The first is the EDA issue you describe. The second is the COTS problem (if this were a really good idea the commercial guys would be all over it, then we in DoD could use it). Finally, even if commercial guys are interested and there are important benefits for DoD, we cannot possibly have enough influence too change the course of the IC industry.

There is truth in all of the above, but we need to see if we can't break out of the Scarlett O'Hara “I'll worry about it tomorrow” mold. I am not an IC designer so don't have lots of “why/why-not” info at my fingertips. But, I am trying to put together a program that makes sense for DARPA to fund. Amongst the questions that will be asked before any money goes out the door are:

  • What are the attributes that will be improved (speed, power, noise, design time etc.) and by how much (2X, 10X, 100X)?
  • What are reasonable interim goals to judge whether effort is on track towards reaching overall objectives, which organizations are most likely to be successful, and why?
  • how much will it cost?

These are questions that I have been grappling with. And since I take help from anywhere I can find it, I am constantly on lookout for people with answers or at least connections to people with answers. We have had workshops with folks from Sun, Fulcrum, Theseus, Intel, IBM, and several universities, but I'm not comfortable that I have the “killer argument.” As your article makes clear, the killer app is obvious, just not the killer justification! Any support/suggestions/references that you can provide to bolster my position would be greatly appreciated.

Bob Reuss
Defense Advanced Research Projects Agency
Microsystems Technology Office
Tactical Technology Office
Arlington, VA

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.