Manufacturing Bits: May 5

Spiking neural network radar chip
Imec has developed what the R&D organization says is the world’s first chip that processes radar signals using a spiking recurrent neural network.

Initially, the chip from Imec is designed for low-power, anti-collision radar systems in drones.

Neural networks are used in the field of machine learning. A subset of AI, machine learning utilizes a neural network to crunch data and identify patterns. It matches certain patterns and learns which of those attributes are important.

Spiking neural networks are different. The goal is to replicate the brain in silicon. The idea is to mimic the way that information is moving in the device using precisely-timed pulses.

Imec’s device mimics the way biological neurons operates in terms of recognizing temporal patterns. It also learns and remembers the temporal patterns.

In addition, the device consumes 100 times less power than traditional implementations with a tenfold reduction in latency. For example, in the device, micro-Doppler radar signatures can be classified using only 30μW of power, according to Imec.

The device was initially designed to support electrocardiogram and speech processing. It can also be reconfigured to process sonar, radar and lidar data.

This in turn makes it ideal for anti-collision systems in drones. Drones require devices with smaller form factors. They need to react quickly to changes.

“Today, we present the world’s first chip that processes radar signals using a recurrent spiking neural network,” says Ilja Ocket, program manager of neuromorphic sensing at Imec.

“Hence, a flagship use-case for our new chip includes the creation of a low-latency, low-power anti-collision system for drones. Doing its processing close to the radar sensor, our chip should enable the radar sensing system to distinguish much more quickly – and accurately – between approaching objects. In turn, this will allow drones to nearly instantaneously react to potentially dangerous situations,” said Ocket. “One scenario we are currently exploring features autonomous drones that depend on their on-board camera and radar sensor systems for in-warehouse navigation, keeping a safe distance from walls and shelves while performing complex tasks. This technology could be used in plenty of other use-cases as well – from robotics scenarios to the deployment of automatic guided vehicles (AGVs) and even health monitoring.”

Kathleen Philips, program director of IoT cognitive sensing at Imec, added: “This chip meets the industry’s demand for extremely low-power neural networks that truly learn from data and enable personalized AI. For its creation, we rallied experts from various disciplines within Imec – from the development of training algorithms and spiking neural network architectures that take neuroscience as a basis, to biomedical and radar signal processing and ultra-low power digital chip design.”

Terahertz data networks
Brown University and Rice University have developed a technology that could pave the way towards future systems with fast terahertz data networks.

Today’s systems can quickly locate and connect to a Wi-Fi network in a given environment. The ability to make that connection is called a link discovery, according to researchers from Brown and Rice.

Going forward, the industry is developing 5G networks in the millimeter-wave range, which relies on a sequential search, according to researchers in Nature Communication, a technology journal.

Terahertz networks are fast and can carry more data than today’s WiFi as well as 4G and 5G networks, but they are not expected to appear in the near future. Plus, terahertz waves propagate differently than microwaves, which means the industry requires a different link discovery technology.

Terahertz waves propagate in narrow beams. The solution? A leaky waveguide antenna. Using a leaky-wave antenna with a broadband transmitter, Brown and Rice have demonstrated a link discovery approach for terahertz radiation. One terahertz is 1,000GHz. Terahertz radiation is also known as sub-millimeter radiation. These are electromagnetic waves within a band of frequencies from 0.3 to 3 terahertz.

A leaky waveguide consists of two metal plates. There is a space between them where radiation can propagate, according to researchers. One of the plates has a narrow slit, enabling some radiation to escape from the unit.

“We input a wide range of terahertz frequencies into this waveguide in a single pulse, and each one leaks out simultaneously at a different angle,” said Yasaman Ghasempour, a graduate student at Rice and co-author on the study. “You can think of it like a rainbow leaking out, with each color represents a unique spectral signature corresponding to an angle.

“It is not just about discovering the link once,” Yasaman said. “In fact, the direction of transmission needs to be continually adjusted as the client moves. Our technique allows for ultra-fast adaptation which is the key to achieving seamless connectivity.”

Daniel Mittleman, a professor in Brown’s School of Engineering, said: “When you’re talking about a network that’s sending out beams, it raises a whole myriad of questions about how you actually build that network. One of those questions is how does an access point, which you can think of as a router, find out where client devices are in order to aim a beam at them. That’s what we’re thinking about here.

“I think some people have assumed that since 5G is somewhat directional, this problem had been solved, but the 5G solution simply isn’t scalable,” Mittleman said. “A whole new idea is needed. This is one of those fundamental protocol pieces that you need to start building terahertz networks.”