Looking beyond CMOS

Some designers and semi-conductor production engineers have questioned the utility of traditional wisdom of the last 30 years that the number of transistors on a chip doubles every two years. I think there is a need to question one more traditional belief: CMOS is superior to all processes.

In the early 1980s it was predicted that GaAs would take over semiconductor manufacturing from CMOS, because GaAs has inherent advantages of speed and lower signal noise. However, an overwhelmingly large number of chips continue to use CMOS technology. IC Insights estimates that of the $140 billion semiconductor market in 2003, GaAs and SiGe processes generated only $2.1 billion in sales. Projected to grow at a CAGR of 16 percent, compound chip market will grow to merely $4.4 billion in 2008, which will be less than 2 percent of global chip sales. This means continued CMOS domination.
I think CMOS has retained its stranglehold because massive investment enhanced its technology to enable its price/performance remain more competitive than that of GaAs and SiGe, which have not drawn significant investments. While CMOS took rapid technological strides, GaAs and SiGe have remained essentially where they were.

However, in nanometrics age, a closer look needs to be taken at non-CMOS processes. Reports are increasingly coming of such processes rescuing designers and production managers unable to get the required performance from CMOS. Implementing CMOS at 90nm node is becoming increasingly difficult. At this node, leakage and power problems, besides capacitance in the wiring that can vary as much as 30 percent, are major factors affecting product performance. Besides, with multi-million-dollar CMOS masks at this node, the economic advantage CMOS has over other processes no more seems certain.






Agere Systems recently used SiGe instead of CMOS for its latest generation of disk drive preamplifier chips. Agere claims that SiGe amplifiers produce much less signal noise than CMOS while using just half the power at the same levels of performance. Agere is so greatly satisfied with the result that a company spokesman is quoted to have said, "We don't believe there's a place for CMOS in preamps."

The one application that will give a great thrust to alternative processes is the cell phone, said to be driving electronics industry growth in the next few years. Cell phone power amplifiers needing high speed find GaAs the best process; and those needing speed as well as power efficiency, SiGe. ADI finds that SiGe is the ideal choice for fiber optic and wireless RF components, while BiCMOS, which gives higher output current, for providing power and performance in high-performance data converters.

The wireless networking chipmaker Atheros Communications claims to be the pioneer company that has managed to build all-CMOS wireless networking chip. Atheros has acknowledged great difficulties in building high-performance RF devices using CMOS, and claims that it will use its success as a "marketing differentiator." I doubt whether the claim will help. Unlike the designer or production engineer, the consumer does not care whether the chip uses CMOS or any other process. He wants a product with the best price/ performance in his application. A possible advantage for Atheros could be that it has multiple foundry suppliers to build the product and so it can bargain on the price. But the scope of such bargain is limited. When foundries are operating at almost full capacity, as of now, bargaining may be difficult.

What is most interesting about the so-called "exotic" processes is that that as they become more popular, their costs are reducing, in contrast with CMOS, which is rising. While some companies have shown that video processing, Wi-Fi, Bluetooth and Gigabit Ethernet devices that combine digital and mixed-signal functions can be designed and built on the same CMOS die, many believe that to create higher quality radios, "exotic" processes hold greater promise than CMOS.


You can reach Kirtimaya Varma at kirti.varma@rbi-asia.com