The transistor, rich in both cutting-edge physics and practical applications, is one of my favorite topics to write about (see BSR issue 17 for my feature article on transistors). So when Intel Corporation, whose best-selling computer chips each contain billions of transistors, announced this month that it had radically reengineered the transistor for its next generation of microprocessors, I immediately knew my next blog topic was in the bag.
My choice of topic was especially easy because Intel’s new transistor design was invented right here at Berkeley, by a group of electrical engineering professors and graduate students in the late 1990s/early 2000s. Back then, microchip manufacturers were worried that their traditional method for improving transistors – by shrinking them – could only continue until the end of the decade, or perhaps slightly longer. Transistor dimensions would reach a point that they simply could not be made any smaller without degrading their functionality.
To Chenming Hu, a professor in Berkeley’s EECS department, the computer industry’s long-term obstacles presented a golden opportunity for someone in academia to jump out ahead of the curve and invent a new type of transistor that could work at much smaller dimensions than conventional transistors. He did some brainstorming - apparently, while aboard a trans-Pacific flight – and drew up an innovative design he called the ‘Fin’ field effect transistor (FinFET). The device’s name refers to the distinctive three-dimensional sliver of silicon that takes the place of the flat slab of silicon used in a conventional transistor. According to Hu’s calculations, by building a transistor around a fin as opposed to a flat surface, it would be possible to reliably operate transistors as small as five nanometers in length. By comparison, today’s transistors are more than 30 nanometers long.
Once Hu recognized the potential of the FinFET design, the next step was to create a prototype. Because the FinFET has a unique three dimensional shape and nanoscale dimensions, this was by no means a trivial task, so Hu enlisted the help of fellow EECS Professors Jeffrey Bokor, Tsu-Jae King-Liu, and Vivek Subramanian. Together with their graduate students, the team came up with a sequence of process steps to build a working FinFET. The most important aspect of their process was that it was very similar to the process for making conventional transistors; it did not require any novel equipment or materials. While this seems like a minor point, commercial chip makers have invested many billions of dollars over a span of decades developing the high-tech machinery used in their fabrication plants, so they would have been unlikely to adopt the FinFET technology if it required a significant overhaul of their existing process flow.
In 2000, the Berkeley team built and tested their first FinFETs and presented their results at a major semiconductor device conference. The rest, as they say, is history. As expected, commercial chip makers were thrilled by this new type of transistor that was amenable to scaling both down in size and up in production quantities. They invested heavily in FinFET development and now, a mere decade after Berkeley presented its first laboratory results, are set to begin shipping them to consumers by the billions.
It appears that Intel will have a head start in bringing FinFETs into production (hence the title of this post). Competitors like the Taiwan Semiconductor Manufacturing Company (TSMC) have indicated that they will not introduce FinFET technology for another few years. Given that FinFETs have a number of technological advantages over conventional transistors, I wouldn’t be surprised if the chip-making industry undergoes a shakeup as a result of their staggered introduction. The earliest adopters will presumably emerge as big winners and sell their wares in coveted markets like mobile devices and data centers. And the losers? Let’s just say you may need to do your shopping here the next time you want to buy their products.