News

Memtransistor-based Collision Detection Scheme is Inspired by Insects

January 25, 2023 by Darshil Patel

Researchers at Pennsylvania State University created an ultra-low power collision detection system inspired by insects such as the locust, which can detect potential crashes, even at night.

There’s a lot that the natural world continues to teach us when it comes to solving a variety of engineering problems. Along these lines, ACS announced that Saptarshi Das, Associate Professor of Engineering, Science & Mechanics at the Pennsylvania State University (PSU) and his colleagues created a collision detection system inspired by insects.

In this article, we explore why collision detection is such a challenge and we examine the details of the PSU researchers’ findings.

 

The PSU researchers build an ultra-low power collision detector, based on insects such as a locust like this. The system can detect potential crashes, even in the dark.

The PSU researchers build an ultra-low power collision detector, based on insects such as a locust like this. The system can detect potential crashes, even in the dark. Image (licensed) from Adobe Stock Images

 

Challenges of Collision Avoidance Systems

As our vehicles become more advanced and autonomous, it poses a challenge for collision detection under poor illumination and at night. Collision avoidance systems (CASs) play an essential role in autonomous and manned vehicles, drones and navigation robots.

Current collision avoidance systems are often complicated and offer insights based on sophisticated enhancement algorithms or thermal sensors, requiring expensive hardware, making them power-hungry and unable to deploy in remote locations.

CASs can automatically brake when another vehicle or any other object gets too close. They can override the driver's control, change the throttle and apply brakes. To detect these objects, they rely on radar or cameras. However, in heavy rain or low light conditions, they use signal processors to analyze the sensor data.

On the other hand, insects like locusts and flies avoid collisions with their reliable spike-based LiDAR (light detection and ranging), which works even at night. Inspired by their highly efficient collision avoidance system, Saptarshi Das, Associate Professor of Engineering, Science and Mechanics at the Pennsylvania State University (PSU, and his colleagues created a spike-based in-sensor collision detection system.

 

Putting Memtransistors to Work

The system uses an optoelectronic IC consisting of atomically thin, light-sensitive memtransistors. They tried to imitate the escape response of lobula giant movement detector (LGMD) neurons found in many insects and demonstrated real-time collision detection at night involving cars.

The team of researchers at the Pennsylvania State University (PSU) designed an algorithm based on neural circuits to avoid obstacles. Instead of processing an entire image from a camera, they selected only one variable: the intensity of headlights. This approach eliminated the need for an onboard camera, making the overall detector smaller and more energy efficient. 

Their system comprises eight photosensitive memtransistors made of Molybdenum Sulfide, which takes only 40 µm2 area and consumes only a few hundred picojoules of energy, thousands of times less than existing systems. In addition, their detector could sense a potential car accident 2-3 seconds before it happened, leaving enough time for the driver to brake.

 

Atomically Thin Solution for CAS

The PSU researchers explored the locusts' LGMD neurons that can perform non-linear computations to ensure the timely detection of head-on collisions. In their paper, they report that spike-based processing allows faster complex tasks with energy efficiencies.

 

Data from the collision avoidance system in insects.

Data from the collision avoidance system in insects. Image used courtesy of Saptarshi Das and co-authors. (Click image to enlarge)

 

Insect-inspired collision detectors based on silicon complementary MOS (Metal Oxide Semiconductor) technology and FPGAs already exist. However, their sensing unit is physically separated from the processor.

Moreover, they don't allow for spike-based information processing. The researchers introduce atomically thin and light-sensitive memtransistors as edge sensors (with in-sensor computing capabilities) to enable low-power collision detection even at night.

A memtransistor is a combination of a memristor and transistor technology. They combine resistive switching with transistor gates to realize non-linear charge transport with control over individual states and switching ratios. In a memristor, multiple memristors can be connected with a single transistor, allowing one accurate modeling of a neuron.

 

Collision Detection Circuit using eight memtransistors.

Collision Detection Circuit using eight memtransistors. Image used courtesy of Saptarshi Das and co-authors. (Click image to enlarge)

 

The new collision detection hardware includes a reconfigurable integrated circuit of eight monolayer Molybdenum Sulfide memtransistors to sense and process the intensity of the car's headlights.

Their design measures only 40 µm2 and consumes only hundreds of picojoules of energy. The team of researchers believe their novel detector can help make existing CASs better and safer.