Four Berkeley faculty members have invented one of the leading candidates for keeping Moore’s Law alive for the next decade.
EECS Professor Ali Javey, Vivek Subramanian, Ali Niknejad, Jeff Bokor, Chenming Hu, and Tsu-Jae King Liu are gathered around an Applied Materials Centura, a tool used to pattern thin films when making integrated circuits. (Photo by Peg Skorpinski) The Fin Field Effect Transistor or "FinFET," as it is called, is an unusual transistor design that can be scaled down to a “gate length” of less than 10 nanometers, about 10,000 times smaller than the width of a human hair. A small gate length—the distance between the source and the drain—shrinks the transistor overall and enables it to switch on and off more quickly and carry a higher current density.
The joint work of Electrical Engineering Professors Tsu-Jae King Liu, Jeff Bokor, Chenming Hu, and Vivek Subramanian, FinFETs grew out of a 1996 DARPA-funded program that produced several other advanced transistor designs. The FinFET was the only design picked up by industry, however, because it could be manufactured without significant modifications to the fabrication process. Several companies, including IBM and Samsung, are in the process of developing FinFETs, and they will soon begin to appear in consumer devices, although not in their most Lilliputian form. As the demand for smaller and faster devices marches on, 20-nanometer transistors are expected to be at the cutting edge of semiconductor technology by the end of this decade, with 10-nanometer transistors coming into production around 2015.
Even five-nanometer FinFETs are feasible, as Yang-Kyu Choi, a Berkeley-trained Korean researcher, recently demonstrated, but this is pretty much the limit. “Below five nanometers, you can have quantum mechanical tunneling and confinement effects,” says King Liu, who is the director of Berkeley's state-of-the-art Microfabrication Laboratory. “That means small variations in manufacturing would result in large changes in performance.”
EECS Professor Tsu-Jae King Liu. (Photo by Peg Skorpinski) King Liu describes the FinFET as “a vertical transistor, kind of like a skyscraper.” Conventional transistors, called MOSFETs (short for metal-oxide-semiconductor field-effect transistors), are fabricated along the surface of a silicon wafer, making them flat and low. FinFETs, by contrast, are fabricated along the sidewalls of a narrow, vertical fin etched into the surface of the wafer. That means they take up far less real estate, enabling a trillion transistors to be packed onto a chip that now holds a mere billion.
The FinFET's vertical design also counters one of the major obstacles in designing very small transistors: current leakage. Compared to a conventional MOSFET, in which the gate perches above a thick (and leaky) substrate of unused silicon, the FinFET's gate straddles the entire fin. This compact design means that no part of the transistor is unaffected by the gate, yielding a sharper, less leaky turnoff at the threshold voltage, as the researchers demonstrated in 1999. It also means that engineers can push the design constraints harder, decreasing the separation between source and drain. To reap the design's benefits, however, the sidewalls of a FinFET must be absolutely vertical, a tricky technical challenge in the fabrication process.
FinFETs still require a lot of computer modeling and simulation to predict how they will behave in an integrated circuit, and King Liu continues to train students to perform these simulations.
Meanwhile, King Liu’s research has taken off in a new direction, away from transistors and toward the challenge of designing low-cost, low-power memories for portable devices such as iPods and Flash memories. “Eight years ago, portable technology wasn’t the research driver, it was high-performance computing," she says. "Today the biggest challenge is to come up with new design concepts to decrease power consumption in portable devices.”
King Liu is working on something she calls a NEMory cell: a nanoelectromechanical memory. Instead of using electronics to store bits of information, a NEMory cell uses physical switches. At first blush, this might seem like a step back into the past, but in fact, King Liu says, there’s a lot to be said for physical switches: They are easier to make and to deposit on a silicon wafer than electronic switches. They can be programmed faster, with lower power, and they are more stable. Moreover, they are immune to radiation effects, and if they are heated up, they don’t leak charge. King Liu has demonstrated that a NEMory cell can be manufactured using standard materials and techniques and integrated with electronic circuits.