Resurrecting The Semiconductor Industry

Every so often the semiconductor industry approaches a wall, and it takes the best minds in the world to figure out whether to go left, right, below, above or through it.
As an industry we faced that at 1 micron, and again at 45nm with lithography. Now we are approaching another wall. Complexity has exploded—hardware, software, IP and process technologies—while the entire time-to-market cycle has been compressed beyond the capabilities of many design teams.

When so many factors conspire together, it’s time to take a step back and rethink where we’re going, how we’re going to get there, and exactly who should do what and for what purpose. These are difficult times and difficult issues, and no one can do them all.

We have been chasing bigger devices that support lots of features and, at the same time, smaller, more expensive feature shrinks to encompass those big devices. The result is that many new products simply can’t be funded because of the price tag, which can be anywhere from $20 million to several hundred million dollars for the chip alone. This is hampering innovation, and limiting the number of new products and new company start-ups. And if it continues, we will destroy the semiconductor industry in our effort to make bigger, more fully featured, faster and power-efficient products.

The rational path forward is to give users what they want, and the obvious approach is simplicity. Rather than trying to cram everything onto the next process technology, why not keep some devices at older geometries and focus on what they do best? If a device doesn’t need to process everything locally, and the emphasis is on cloud computing, in many respects these become dumb displays with fast I/O and sophisticated network connectivity. Devices can also be composed of standard processor subsystems, combined with readily available IP (e.g. a multimedia core, security core, standard interfaces, etc), and differentiated with considerable software content. This gives the smaller companies a quick and less-expensive, less-risky path to a product.

The IP in these products can be standard, so time to market and IP integration will be quick, and semiconductor companies can partner closely with EDA and a few trusted IP providers up front to ensure the platform they want to build can be created at the lowest possible cost.

For those applications that do require advanced processing power—the chips can be made small enough and specific enough to be used as platforms that either run alone or in stacked die packages—they can still push the limits of physics and Moore’s Law. But the upside is this won’t limit the whole semiconductor market. It will be a gamble of a few large companies with deep pockets and the internal resources to build the very best processor or memory architectures for markets where there is either enough volume to warrant it or enough price elasticity to pay for it.

We are shifting from a world of complexity to one of complex interactions using simpler parts, where the key will be in architecting platforms that use resources rationally. Instead of just shrinking features, we’re expanding the reach of technology, and from that point onward the stranglehold that growing complexity has on the market will be reduced. That, in turn, will result in more design starts, more investment in the industry, and an overall healthier semiconductor industry. Design starts will increase and investor dollars will begin rolling back into the semiconductor industry.

Change is upon us. The question now is who will grasp what amounts to a seismic shift and take advantage of it.