ICs do a lot, but circuit design is still a needed art
In the past few years, I have felt—and told many people—that the age of circuit design is over. By "circuit design", I don’t mean IC design: I mean taking ICs, discrete devices, and passive components, then combining and configuring them to create the distinctive topology which addresses multiple design issues, while meeting an application's requirements.
Why have I felt this way? Because it seems to me that the circuit-design challenge and battle has been largely, although not entirely, won by the IC vendors. That's the case even in the subtle areas of data acquisition and test/instrumentation, where the analog front end must be carefully tailored to the transducer as well as operating conditions. Much of circuit design now consists mostly of selecting the right ICs, being sure that they interface properly with each other and the I/O, and then providing software to make them execute their roles properly.
For example, you can now get high-performance op amps and instrumentation amps (in-amps) for modest cost. No longer do you need to select passives and carefully match temperature coefficients or worry about drift. In many cases, they are now embedded in the IC and thus inherently matched, or the IC has some clever scheme that cancels out many drift errors. Further, vendors offer reference designs and development kits which feature tested and verified circuits for many common, and even some unusual applications.
But as with so many of my pronouncements, I was wrong. Here's why: I recently saw a design and actual prototype done by Jim Williams and Omar Sanchez-Felipe of Linear Technology Corp. Jim's formal title is Senior Scientist, but that's somewhat misleading: he is among the preeminent analog-circuit designers in the world, with a career which spans 40+ years. I still marvel at his first major published design for a portable scale for the MIT nutrition lab, which had to resolve to 0.1 oz, never need calibration, and be built using only standard, off-the-shelf components. It was—and still is—a marvelous example of understanding every subtle source of signal-chain error and then figuring out how to minimize or cancel each one.
What have Jim and Omar done? They started with what seems to be a simple, straightforward way to measure temperature called acoustic thermometry, which is based on the consistent variation in the speed of sound as temperature changes. In principle, this should be an easy circuit: all you need is to take an ultrasonic transducer, pulse it, measure the transit time of the echo, do a simple calculation, and you are all done. What's the big deal?
But principle is not reality; it's not even close to it. The final circuit required careful management of low and high voltages, pulse timing, signal lockout, and many other factors, plus physical placement issues. You can read the full details in their Application Note 131, An Introduction to Acoustic Thermometry here, or an similar version here. While you are doing that, I will use this application and its physical realization in an actual circuit design as a constant reminder that there still is room for what we understand in our gut as The Existential Pleasure of Engineering (to use the title of Samuel C. Florman's excellent book). ♦