Engineers at Purdue University and GlobalFoundries have gotten today’s most advanced transistors to vibrate at frequencies that could make 5G phones and other gadgets smaller and more energy efficient. The feat could also improve CPU clocks, make wearable radars, and one day form the basis of a new kind of computing. They presented their results today at the IEEE International Solid-States Circuits Conference, in San Francisco.
5G transceivers need two key components: oscillators and filters. Oscillators produce the frequency signal that is then modulated to carry data, while the filters allow the transceiver to tune in to specific channels. Both need components that can resonate at gigahertz frequencies.
Usually, to get such a high and stable frequency, you need a crystal oscillator, a device that resides outside the processor. Its megahertz-scale signal is then multiplied in successive stages by circuits called phase-locked loops to climb into the gigahertz region. The signal is then distributed around the chip. The setup consumes more power than designers would like, and because it has an extra component, it takes up more space than you’d want.
“You’d get much better performance if you start with a good oscillator at 32 GHz instead of multiplying,” says Bichoy Bahr, who worked on the resonator technology at Purdue while pursuing his PhD in the laboratory of Dana Weinstein. For that, you want a component—ideally something tiny that can be integrated on the same silicon as the radio chip—that resonates at the frequency you want.
Taking their cue from tuning forks, engineers have tried integrating resonators using microelectromechanical systems. These have components tuned to physically vibrate at a particular frequency and produce a useable electronic signal. But they have limitations. Most require different processing than the CMOS chip they’re on, so they have been built either atop the finished chip or embedded inside the wiring layers above the silicon.
Weinstein’s group instead found a way to make a MEMS resonator using the CMOS transistors themselves. Leading-edge chips today rely on FinFETs, transistors where the current flows through a protruding 3D fin of silicon. Current is controlled through a gate electrode that drapes over the fin but is separated from it by a thin layer of dielectric.
That gate structure creates a capacitor, so when voltage is applied to the gate electrostatic forces squeeze the dielectric. A regular series of voltage pulses will then create a periodic acoustic pulsing in the transistor. “We’re pulling and pushing on the fin width,” says Weinstein. By spacing a series of fins in just the right way and connecting them all with the same gate the whole structure resonates at 32 GHz.
By building a set of copper planes in the wiring layer above the FinFETs, this 5G fin vibration is confined to just the 154-transistor structure. So even if it’s integrated into a radio chip, the vibrations won’t leak out to affect other circuits.
The periodic fin-squeezing leads to detectable changes in the current that flows through the transistors, so by tapping a subset of the 154-fins, Weinstein’s group could pull out a clean 32-GHz signal. In fact, the signal was so good it surpasses any other room-temperature silicon MEMS-based resonator to date. They measured a quality factor (a dimensionless quantity that tells you how awesome your resonator is) of nearly 50,000. “You get all of this in a micron square or less of silicon,” Weinstein says.
Importantly, the structure was built entirely using chipmaker GlobalFoundries’ 14-nm CMOS production process. So integration into a real radio chip should be straightforward. (GlobalFoundries engineer Zoran Krivokapic was part of the team.)
But the resonator is just the main ingredient in oscillators and filters, so Weinstein’s team is focused on building those complete circuits. Still, they already have their eyes on applications. Aside from 5G radios, the resonant fin structure could improve processor clocks. First, it could replace the off-chip oscillator with a small sliver of a processor’s own silicon. Second, it could potentially distribute the clock signal around the chip better. The resonator, which is just 200 nanometers wide, acts like a waveguide, losing very little of its acoustic energy along its length. Weinstein imagines clock signals being distributed to distant parts of the chip using low-loss acoustic waves instead of electronic signals; about one-third of a processor’s energy is today taken up in distributing the clock signal.
What’s more, Moore’s Law will naturally increase the frequencies resonant fins can attain. Looking ahead to GlobalFoundries 7-nm process, the team figures 60–80 GHz is attainable. That puts the resonators in the realm of car radar. With such an efficient radar source at hand, Weinstein envisions wearables that do personal navigation.
As important as the resonator technology is to near-term applications, it was developed under a pretty futuristic mandate. The Defense Advanced Research Agency’s UPSIDE program provided the funding. Its goal is to find ways of analyzing image data using the physics of oscillators rather than their digital nature.