(Author's note: This column originally appeared in EETimes , December 15, 2008.)
As the industry enters one of its periodic downturns, the feeling I get is variation on a line from the classic film Casablanca: “but we'll always have analog.” Vendors are responding to the eternal nature of analog functions and principles, defined by both Moore's law (actually, Moore's conjecture) and the even stronger laws of physics, and adjusting their strategies.
Why? Because analog thrives, and is increasingly needed, as a foundational, enabling technology and function. Analog, in the broad sense, deals with five fundamentals: signal input/output (sensors and transducers); power supplies; interfacing (level shifting, drivers/receivers); signal processing over entire electromagnetic spectrum; and signal integrity. As digital systems increase in speed and density, these analog issues don't go away; they become more challenging, more critical, and less forgiving.
Despite the opportunities presented by Moore's law, and relentless pressure it produces on IC design and fab, “analog” is an ever-more vital part of the circuit, system, and even software engineer's concerns. And every time the industry gets a cold, vendors with strong analog capability are in a better situation compared to digital vendors, while those with only slight or no analog capability realize that analog holds better business potential, and so they start to get on with (or even resume their connection to) the analog world.
From a business perspective, the model for the longevity and potential profitability of analog (including power-related) ICs are due to different design imperatives, selling cycles, product lifetimes, and customer perspective than the all-digital world. Unlike digital ICs, where density is key and it's mostly about die size and cost per wafer, analog ICs tend to be focused on fewer active devices per die, with die size just one of many factors defining cost.
More importantly, analog ICs tend to have a different “value proposition” to their users. A circuit designer may be willing to spend another 50 cents or dollar for a single, high-precision instrumentation op amp with a lower output-drift temperature coefficient, yet this op amp has relatively large die size and few active devices. Why spend the money? Because the better-performing device can eliminate an assembly-line calibration cycle or performance issues in the field, both of which cost money and headaches, even if it is not obvious from the BOM alone.
Further, the functions which analog provides within many systems tend to be relatively fixed from product cycle to cycle, so even as the more glamorous digital and software portion upgrades relentlessly in our brutally short product cycles (“now with more memory and more features!”) there is less need to change the analog side. Ironically, in fact, there is reluctance on the part of the designer to do so: it's better and smarter to leave some of the product design unchanged, to reduce how many new subcircuits have to be juggled and debugged. Thus, the analog parts have a much longer viable design-in life, and so provide IC OEMs a better opportunity to get some real ROI, especially since the manufacturing margin on an IC increases over with time and experience.
IC vendors know that analog markets are more defensible at the customer design level. That's why the top-tier vendors are pushing investment in analog process, design, tools, and people. And it's why we see vendors with little or no analog expertise try to get in on the act, adding some basic analog functions to their portfolio, in the hope of capturing even a small slice of their customers' needs.
I suspect the smarter digital vendors also have another worry in the back of their minds. They know that Moore's law can't go on forever; nothing does. At the same time, the demands for even more functionality per digital IC is pushing the design challenge, fab cost, and power limits perhaps to an unsustainable point.
A new approach will be needed, and it may turn out to be the old approach of doing it in analog, which can be more effective against those three limits. After all, an analog filter costs only a few op amps, and consumes little power, compared to a digital version implemented as hardware or in a programmable device (see National Semiconductor's LMV1088, a dual-input, far-field noise-suppression microphone amplifier, or the work of MIT's Analog VLSI and Biological Systems Group, http://www.rle.mit.edu/avbs/).
Maybe the future of ICs is one of increasingly hybridized monolithic devices, where analog signal processing takes on some of the formerly digital functions, and digital handles only those where it provides a clear performance or cost advantage. ◊