Startup Wants Better IoT Radios

By Rick Merritt, Cupertino, CA

The Internet of Things needs new lower power radios and a fresh approach to silicon design, said an expert with a startup designing IoT chips.

“I think there’s a growing realization Bluetooth Low Energy and Zigbee are not the right solutions” because they consume tens of milliwatts, said Benton Calhoun, co-founder of PsiKick (Charlottesville, Virginia). “People are talking about new solutions, but it’s not clear what they are yet,” he told an audience of several hundred silicon designers at the annual Hot Chips event here.

For its part, PsiKick is designing an ultrawideband transceiver into an upcoming IoT node chip geared to consume a total of less than 30 microwatts. But UWB comes with a number of tradeoffs in potential interference, lack of compatibility and relatively low data rates.

“It’s possible transmit 200 Kbits/s at 4 microwatts and Mbit/s at 30 microwatts at one meter with UWB, but there’s no receiver in your phone for it,” Calhoun said. New IoT radio standards “that make more sense than ones used now” will emerge from a range of proprietary radios, he said.

Radios are just one piece of the broad need for lower power chips, said Calhoun who conducts research in the area at the University of Virginia. “Today’s devices won’t get us to trillions of nodes people predict for IoT,” he said.

Some market watchers say a wealth of microcontroller-class SoCs are available for IoT. Calhoun argues they run at tens of milliwatts, requiring batteries that eventually need to be replaced.

“Doing 275 million battery changes a day cannot be the solution for a trillion-node IoT network — the answer is energy harvesting,” he said, admitting harvesters will constrain silicon to tens of microwatts.

PsiKick’s upcoming SoC will integrate an ARM Cortex core along with a number of blocks the group has already designed in academic research. It also includes a bi-directional UWB transceiver and interfaces to analog and digital sensors as well as support for a variety of harvesters.

Today’s ARM cores also consume too much power for IoT, “but the market wants ARM,” Calhoun told EE Times, admitting emerging alternatives such as the open-source RISC-V core consume less power. “System integration is the hardest part of these designs because if you’re not careful about the blocks you use you quickly get back up to tens of milliwatts,” he said.

Another drawback to the ultra-low power target is some things have to be left out such as security hardware. “We know public key encryption consumes too much power for IoT, but we’re waiting to see what [IoT security] standards develop,” he said.

A principal engineer from Intel Labs agreed with some of Calhoun’s points in a separate talk about a research chip it designed. SoCs supporting always-on audio and video — applications expected to be in high demand in future wearable devices — would need to recharge every hour or two, said May Wu, a director of Intel’s wireless lab.

“The most interesting wearables are those that include audio and video in devices that evaluate what to see and hear,” Wu said. But there’s “a 20-fold gap between what’s needed and the power consumption of current devices,” she said.

Intel Labs’ SoC for wearables uses a version of the Currie x86 core.
Intel Labs’ SoC for wearables uses a version of the Currie x86 core.
In an effort to explore trade-offs for next-generation wearable SoCs, Intel designed 14nm test chips with specialized audio and video blocks. The chips use a version of the x86 core in Intel’s Curie module along with a microcontroller to handle light jobs when the host is turned off.

Researchers are still working on the algorithms for the SoCs aimed at uses such as speech and gesture recognition. They are also investigating future applications for them as well as, like PsiKick, new low-power radios for them.