Think IoT Designs Are Challenging? Try Embedded Systems In The Brain

A change of just 1 degree Celsius can cause permanent damage.


There’s low power and then there’s low power. There are amazing applications and then there are amazing applications.

Today the bleeding edge of low power design is not so much in IoT (although excellent work is being done in that space) but in medical, where the stakes are high and possible outcomes life-altering.

Chet Moritz, associate professor with the University of Washington’s Departments of Rehabilitation Medicine and Physiology & Biophysics, and his colleague, Visvesh Sathe, assistant professor of electrical engineering at UW, are two researchers walking that bleeding edge. They and colleagues at UW’s Center for Sensorimotor Neural Engineering and partner universities around the country are working to create embedded systems to restore movement to paralyzed patients.

“We are trying to connect the human brain to computer chips,” Moritz says.

They’ve eyed four grand challenges:

• Decode movement intention from neural recordings
• Restore movement
• Interface with the adaptive brain to restore sensation
• Induce plasticity for recovery

Medical researchers have been working on these problems—in one form or another—for a long time. One big stumbling block of late has been that the technology systems are large, slow and essentially bifurcated between inside the body and outside. For example, a series of probes are attached to the brain to capture neural sensation or intention and these are threaded outside to external devices, which in turn feed back stimulus signals to affected parts of the body. This of course is cumbersome, costly and slow, but it’s progress.

The prize is an embedded system. That should come as no surprise. And also what should come as no surprise is achieving that will require extraordinary engineering and design effort in the coming years.

Sathe said the mission is to build an implantable bidirectional brain-computer interface.

“It’s implantable so you go through surgery, you implant the chip into the brain along with stimulation or sense electrodes and then it gets sealed up,” he said. “Importantly it gets sealed up, so there are no wires hanging out … otherwise there’s risk of infection.”

With masterful understatement, Sathe, an AMD veteran, noted that “like any system it has its own constraints.”

Let’s look at just a couple of them.

This is nothing new to system designers, especially in the mobile and IoT spaces, where performance throttling sometimes occurs to prevent thermal runaway. In such devices, the consequences are hot hands or malfunctioning systems. If the embedded system inside your brain is too hot, the consequences can be dire.

“Far from a heat sink, we can’t go beyond 1 degree Celsius because it’ll cause damage,” Sathe said. “We’re basically stuck with doing whatever we want to in terms of computation, sensing, and stimulation and yet we can’t go over 1 degree Celsius.”

Talk about constraints.

He describes a system in which sense electrodes detect either neuron activity or “more aggregate local field potentials corresponding to multiple neurons,” push them onto the chip, which communicates to a rack-mounted server for clinical applications. The system also could implement computation in a closed-loop control format and then go back and stimulate activity in the brain or spinal cord, he added.

Real estate concerns
The system requires 256 to 512 electrodes with a minimum sample rate of at least 30 kS/second. That requires at least a 14-bit ADC to digitize and process the information with an aggregate bandwidth of 215 Mbps to communicate to the server, Sathe said.

Size, naturally, matters. Since it needs to be implantable, the IC solution must be around 10x10x0.5 mm, he said. This type of size constraint eliminates the possibility of an onboard battery. Energy harvesting techniques become a challenge because that usually requires super capacitors, and “That’s not an option either,” Sathe said.

Instead, researchers are looking at wireless power transfer schemes. He cited work by his colleague Joshua Smith at UW’s Sensor Systems Lab, where researchers are trying to perfect wireless resonant energy link (WREL) technology. Smith and his colleagues recently prototyped a 130nm high frequency power harvester and regulator in silicon enabling 13.6mhz frequency for power transfer, Sathe said.

These are just a couple of the astonishing challenges facing researchers in the field of neural engineering. I’m not doing this justice because what Sathe, Mortiz, Smith and others are attacking is an enormously complex set of engineering problems, but that’s what puts the sparkle in engineers’ eyes isn’t it?

They’re pushing the very edge of engineering and ultimately their breakthroughs will feed back into other application spaces, benefitting others as well.

It occurred to me as I listened to a presentation from Moritz and Sathe recently that something that was impossible when I was born very likely will be made possible in my lifetime.