Pooh on sloppy CMOS op amps. You know, the noisy ones with really feeble gain-bandwidth products. The kind of op amp with low-voltage, single-supply operation as its only claim to fame. One meant to satisfy some vague "analog" requirement on an otherwise humongous CMOS system-on-chip. Try driving a laser trunk line with that.

The truth is, if you need to push dense wavelength-division multiplexing (DWDM) over 80 km of fiber-optic cable, you need real analog expertise. And if you don't have it in-house, you buy it-or steal it. And increasingly, Linear Technology Corp. is the source.

About two years ago, Xicor Corp., long known as a nonvolatile (E2PROM) memory company, saw an opportunity in laser control modules. The temperature of an optical laser, especially the kind that drives long-distance trunk lines, must be carefully controlled to facilitate phase alignment with receivers and to prolong its life. DWDM lasers, which multiplex additional communications channels on different laser colors (that is, with different frequencies and wavelengths) are even more sensitive to temperature. This is invariably a function of drive current.

Xicor discovered that a multiplying D/A converter (sometimes called a "digital potentiometer") can regulate the current through the laser-and its own E2PROMs could establish the temperature-control set point for the D/A.

Suddenly, Xicor was an analog company. It published a road map showing not only the digital potentiometers but a suite of integrated temperature-monitoring and laser-drive circuits. It hired Lou DiNardo, Linear Technology's marketing vice president, to be president. DiNardo did interviews in which he suggested the company's revenues from memory were flat to dwindling, while its analog revenues were going up and up.

But LTC did not sit on its duff. It just introduced a temperature-control module for DWDM lasers that's accurate to within 0.1 degrees C. The module design, which credits staff scientist Jim Williams, resembles a linear voltage-regulator circuit. An instrumentation amplifier serves as the error amp to the module, increasing the sensitivity of the temperature-control loop down to 0.01 degrees C. But the regulator itself is a switching device, which drives four MOSFETs and a thermoelectric cooler in an H-bridge configuration.

Try doing that as a SoC in sloppy CMOS. I dare you.