The International Solid State Circuits Conference (ISSCC) was held recently in San Francisco. The conference highlighted the many advances in circuit design along with a broad sprinkling of diverse topics. The plenary sessions included a talk from Naoki Noda of Japan's Toyota Motor Corp. on the semiconductor content in 21st century cars. He talked of the "highly intelligent" car that will interface to the "smart highway." Guenter Weinberger of Infineon Technologies (Munich, Germany) talked of ubiquitous wireless connectivity. Finally, Mark Pinto of Lucent Technologies (Allentown, Pa.) talked about atomic-level ICs.
All of the speakers described next-generation circuitry that included analog, mixed-signal and RF components in addition to the massive number of logic gates, processors and memory in the designs. Today, all nondigital designers--including RF, small signal, and power--make up less than 15 percent of the design community. Currently, a system-level chip with millions of gates might have only one analog designer working on a small section of the circuit. This sole designer could take as long to design 1,000 transistors as the rest of the team takes to put together 10 million gates. In addition, the analog designer will take up a very large amount of the silicon space and impose very difficult layout constraints on the surrounding circuitry on the IC.
If, indeed, the next generation of systems includes much more nondigital circuitry, where then will all of the designs come from? The AMS extensions to Verilog and VHDL are good for some of the general-purpose analog and mixed-signal functions, similar to calling up a packaged standard product for a function. Unfortunately, interfacing to sensors requires great care in controlling sources of errors like noise, drift, temperature coefficients, sample-and-hold aperture and jitter, and nonlinear responses. This is more complex than just connecting some amplification and an analog-to-digital converter to the system. The greater the diversity of inputs to control, the wider the range of the circuitry to measure the signals. This problem is on an even greater scale for the RF inputs, which are looking for microvolt-level signals that may be rapidly changing frequencies as a part of the digital modulation supposedly yielding better noise rejection.
At the other end of the system, the power circuitry for the actuators could create enough noise and power surges to disrupt even the digital logic. Again, this isn't just a question of finding the largest driver and connecting it to the system. It requires careful control of the interfaces and circuits. The RF power amplifier has similar requirements in that it must meet the frequency accuracy and settling-time requirements while powering up the passive matching components.
Just as hardware people "let" the software fix the interface issues in the systems integration, the digital people go to the analog and interface people when the hardware doesn't work. The problems may be in the digital sections--as in a bad handshake or inverted-control signal--but the fix is executed in the nondigital sections or in the software. These portions of the design are in another domain and therefore fair game.
So where will we get the people with this type of design experience? Most of them retired way back at the end of the cold war when the military suppliers started consolidating and going out of the analog and mixed-signal business. How many engineers do you know who have analog IC design experience? How many RF designers are there in your company? The next-generation ICs will need designers with IC design experience and also a fair understanding of the whole system--the system-level requirements will take extraordinary precedence over device-level requirements.
We can't count on the colleges. They can't teach what people aren't interested in learning and, furthermore, can't teach industry and system-level design experience. Not to sound unfairly biased, but would you entrust the success of your whole system on a chip to an inexperienced recent college graduate? When the mask set costs over a million dollars, you don't have the luxury of planning for a silicon spin. Everything must work perfectly on the first pass.
-This editorial originally appeared as an "Editor's Notebook" entry in ISD Magazine, February 2000.