Today’s wireless technologies harness chunks of the electromagnetic spectrum for myriad uses — radio waves broadcast TV and radio; microwaves transmit cellphone signals and cook our food; X-rays image our bodies; gamma rays kill cancerous cells.
Hossein Naghavi, however, is interested in more neglected slices of the spectrum. Naghavi, an assistant professor of electrical and computer engineering at the University of Washington, studies the “terahertz band,” a region of the spectrum sandwiched between microwaves and infrared waves. Terahertz frequencies are notoriously difficult to work with, but they hold enormous potential in the fields of sensing, imaging and communications — future sensors, for example, could help firefighters “see” through smoke during rescue operations.
Naghavi recently joined a cohort of researchers from across the country who were awarded grants by the U.S. Department of Energy’s Genesis Mission, an initiative to apply artificial intelligence across a wide range of research areas; other UW researchers are part of a Genesis-funded project to advance AI-driven cosmology. With the grant, Naghavi plans to develop compact, efficient sensors that could enable wearable gadgets to image their environment in new ways.
UW News caught up with Naghavi to learn about his new project and how it extends his work on terahertz frequencies.
What is the terahertz band and why are you studying it?
Hossen Naghavi: The terahertz band is a segment of the electromagnetic spectrum that lies between 100 gigahertz and 10 terahertz — the microwave band sits below it, and the optical band sits above it. That position gives terahertz waves a unique combination of microwave and optical properties. Microwaves can see through opaque materials like clothing, smoke or fire, but their long wavelengths limit the resolution of microwave imaging. Optical waves have the opposite problem. Their wavelengths are short, so they produce high-resolution images, but most materials block visible light completely, which makes it impossible to see inside or behind an object.
Terahertz waves are a sort of “happy medium.” Their wavelengths are short enough to give useful resolution but long enough to see through many materials. That combination allows us to build new sensors and cameras that can detect concealed objects or image scenes through smoke, dust and other conditions that defeat conventional optics.
What are some applications you envision for terahertz frequencies?
HN: Augmented reality is expected to become a defining mode of human-computer interaction, but realizing its full potential requires machines that can perceive and understand their surroundings far beyond what the human eye can see. Consider a high-stakes setting such as firefighting, where an augmented reality headset powered by terahertz waves could help firefighters locate victims or identify hazardous materials through smoke, fog and debris.
Beyond firefighting and emergency response, terahertz technologies could also aid in autonomous navigation, security screening, industrial inspection, biomedical sensing, molecular spectroscopy, agricultural applications, and 5G and 6G communication networks.
Sounds exciting! What’s the catch?
HN: Sensors that use terahertz waves, like the ones in our firefighting headset example, have been demonstrated in the lab. However, low-cost, low-power electronics that would be practical in a wearable device have not yet been developed.
Terahertz sensors produce high-resolution image streams, and processing them conventionally means moving enormous amounts of data to a central processor for analysis by an artificial intelligence system. That consumes too much power and adds too much delay to be practical in a lightweight device meant to be worn all day.
Tell us about your new project. How will it address some of the hurdles facing terahertz technologies?
HN: The usual way to build a terahertz imager is to split the job in two. The radar sensor collects raw signals, and a separate processor turns the signals into a picture. That division sounds sensible, but it is the source of most of the trouble. The raw signals arriving at each of the sensor’s antennas are slightly out of step with one another, and the processor has to line them all up before an image can form. That alignment requires a lot of continuous computation, which drains batteries quickly and introduces lag.
Related
Read more about Hossein Naghavi in this profile from the electrical and computer engineering department.
What we are proposing is to stop treating sensing and computing as two separate steps. Instead of collecting raw signals and fixing them afterward, our sensor does the aligning as it collects. We add tiny analog memory cells throughout the sensor which adjust the signal on the fly, as well as an artificial intelligence layer that supervises those adjustments as conditions change. The result is that the signal comes out of the sensor already organized. Very little raw data ever has to leave the chip because the sensor both sees and thinks.
The natural comparison is the human eye. Your retina does not ship every photon to your brain for interpretation. It processes what it sees on the spot and passes along something much more compact, which is part of why vision costs your body so little energy. We are trying to give a terahertz sensor the same quality, which is why we describe the design as “neuromorphic,” meaning “brain-inspired.”
Who are you working with on this technology, and what’s next?
HN: My group at the UW and Milad Koohi‘s group at Texas A&M University are designing and building the sensor hardware. Morteza Fayazi at the University of Utah and Dan Elmhurst at ChipNexus are developing and implementing the AI system. This is a highly collaborative project.
Our next big milestone is to demonstrate a terahertz neuromorphic imager as a proof of concept in Phase I of our Genesis Mission project. Moving forward, we hope to expand the project into Phase II to add even more capabilities and make this technology accessible for public usage as early as possible.
For more information, contact Naghavi at naghavi@uw.edu.