原创 A look at the 4004, nanometre CMOS VLSI and MEMS logic (Part 2)

[Continued from A look at the 4004, nanometre CMOS VLSI and MEMS logic (Part 1)]

Synchronous and asynchronous MOS VLSI logic circuit methodologies are now in the mainstream of electronics design. But the end is near for this approach: five or ten years at most, by most knowledgeable projections. And while further integration is possible, the cost of doing so is escalating: it now takes a company with the net worth of a small country to afford the cost of fabricating today's nanometre circuits.

Intel Corp., long an advocate of the synchronous VLSI circuit design, recently faced the fact that it is not possible to achieve higher levels of integration with the traditional planar MOS approach that has worked so well for 30 or 40 years. So it is now pouring billions of dollars into going vertical with 3-D FinFETs to achieve higher density, and more importantly, lower power. It also recently acquired Fulcrum Microsystems, whose main claim to fame was the low power and blazingly fast networking devices it built using a proprietary asynchronous logic based on technology originally developed at Caltech.

So, what next? There are quite a few alternatives being proposed: nanoscale logic built from carbon-based Fullerene structures, nanowires, DNA nanoscale structures structured to operate in an on/off form similar to silicon-based logic – and, of course MEMS logic. This last is the alternative that I favour, partly because I have always been attracted to the elegant "no nonsense" simplicity of all of Dr. Feynman's ways of thinking as reflected in his ideas and theories.

Working in its favour is that MEMS are already used as a complement to the CMOS logic in many consumer and embedded devices to replace mechanical functions such as gyroscopes and motion sensing in a range of automotive, industrial and mobile communications devices not amendable to scaling. Already in existence are a number of design tools from the likes of Mathworks, Cadence and Coventor to help in the fabrication of MEMS sensors and other miniaturized mechanical components.

Using this as a starting point – and possibly attracted as I was by the elegant simplicity of the approach – there are a growing number of researchers looking at the feasibility of VLSI scale MEMS logic. Since the year 2000 virtually every Integrated Solid Circuits Conference (ISSCC) has had one or more papers on ways to build MEMS logic devices.

A search of papers on ACM, IEEE and Google Scholar on the concept has turned up some interesting results. In a paper presented at the 2010 ISSCC (pdf), a team of researchers from MIT, UC Berkeley, and UCLA demonstrated the feasibility of building a wide range of integrated MEM logic circuits, including switches, inverters, carry-generation circuits, flip-flops and latches. They also created a digital-to-analogue converter (DAC) from MEMS logic structures, as well as a 10bit DRAM column composed of MEMS switches.

And this year, in a paper in IEEE Transactions on Electron Devices (pdf), a team of researchers from UC Berkeley analysed the feasibility of building and scaling MEMS switches and relays for use in ultra-low power digital logic. One of their conclusions was that scalable MEMS logic technology could make it possible to achieve a 10-fold increase in energy efficiency as compared to equivalent CMOS VLSI technology for circuits operating at clock frequencies of about 100MHz. While not up to the bleeding edge performance necessary in many PC and mobile designs, this is more than enough to interest many embedded developers.

Despite all of the barriers to MEMs logic ever replacing MOS VLSI, think about some of the advantages, even if only as a niche or complementary technology to mainstream semiconductor logic. For one thing, compared with semiconductor-based circuits, MEMS-based logic devices would be virtually immune to most harsh environmental conditions, including the radiation environments in outer space.

Second, a MEMS logic structure is absolutely nonvolatile. If the power goes out, there is no need to quickly store valuable information in nonvolatile storage. Power outages would have no impact: the logic state it was in before would be the logic state it would be in after the power is restored. Because of its nonvolatile nature a MEMS-logic based computer would require very little power when compared to existing systems where there is always power required to maintain the logical state of the system.

Realistically, I expect that even if VLSI level MEMS-based logic devices were possible, institutional and educational momentum would work against their widespread use. But darn it, I have liked the elegant simplicity of the approach ever since I heard Dr. Feynman talk about it (but that is another story). And nothing I have learned about semiconductor electronics since then has changed that. What do you think?