AI Efficiency Skyrockets by 100 Times

Evanston, Ill. – Forget the cloud.

Engineers at Northwestern University have developed a new nanoelectronic device that can perform precise machine-learning classification tasks in the most energy-efficient way yet. Using 100 times less energy than current technologies, the device can crunch large amounts of data and perform artificial intelligence (AI) tasks in real-time without sending data to the cloud for analysis.

With its small footprint, extremely low power consumption and lack of lag time to obtain analysis, this device is ideal for being incorporated directly into wearable electronics (such as smart watches and fitness trackers) for real-time data processing and near-instant diagnostics. Ideal for.

To test the concept, engineers used the device to classify large amounts of information from publicly available electrocardiogram (ECG) datasets. Not only can this device efficiently and accurately identify irregular heartbeats, but it is also able to determine arrhythmia subtypes out of six different categories with approximately 95% accuracy.

This research will be published in the Nature Electronics journal on October 12.

“Today, most sensors collect data and then send it to the cloud, where analysis occurs on energy-hungry servers before the results are ultimately sent back to the user,” Northwestern said. Mark C. Hersam, senior author of the study. “This approach is incredibly expensive, consumes significant energy and has time delays. Our device is so energy efficient that it can be deployed directly into wearable electronics for real-time detection and data processing “, which may enable more rapid intervention for health emergencies.”

A nanotechnology expert, Hersam is the Walter P. Murphy Professor of Materials Science and Engineering at Northwestern McCormick School of Engineering, He is also the Chairman, Director of the Department of Materials Science and Engineering Materials Research Science and Engineering Center and members of International Institute of Nanotechnology, Hersam co-led the research with University of Southern California professor Han Wang and Northwestern research assistant professor Vinod Sangwan.

Before machine-learning tools can analyze new data, these tools must first accurately and reliably sort the training data into different categories. For example, if a device is sorting photos by color, it must recognize which photos are red, yellow, or blue in order to accurately classify them. An easy task for a human, yes, but a complex – and energy-hungry – task for a machine.

For current silicon-based technologies to classify data from large sets like ECG, it takes more than 100 transistors – each requiring its own energy to run. But Northwestern’s nanoelectronics device can do the same machine-learning classification with just two devices. By reducing the number of devices, the researchers significantly reduced power consumption and developed a very small device that can be integrated into a standard wearable gadget.

The secret behind the new instrument is its unprecedented tunability, which results from the blend of materials. While traditional technologies use silicon, the researchers created the miniature transistor from two-dimensional molybdenum disulfide and one-dimensional carbon nanotubes. So instead of requiring multiple silicon transistors – one for each stage of data processing – reconfigurable transistors are dynamic enough to switch between different stages.

“The integration of two different materials into one device allows us to tightly control current flow with applied voltage, enabling dynamic reconfiguration,” Hersam said. “Having a high level of tunability in a single device allows us to execute sophisticated classification algorithms with a smaller footprint and lower energy consumption.”

To test the device, researchers looked at publicly available medical datasets. They first trained the device to interpret data from ECGs, a task that typically requires trained health care workers significant time. Then, they asked the device to classify six types of heartbeats: normal, atrial premature beat, premature ventricular contraction, paced beat, left bundle branch block beat and right bundle branch block beat.

The nanoelectronic device was able to accurately identify each arrhythmia type in 10,000 ECG samples. By bypassing the need to send data to the cloud, the device not only saves significant time for the patient but also protects privacy.

“Every time data is moved around, the potential for data theft increases,” Hersam said. “If personal health data is processed locally – such as on your wrist in your watch – it presents a much lower security risk. In this way, our device improves privacy and reduces the risk of a breach. Does.”

Hersam envisions that, eventually, these nanoelectronic devices could be incorporated into everyday wearable devices, personalized to each user’s health profile for real-time applications. They will enable people to make the most of the data already collected without consuming electricity.

“Artificial intelligence tools are consuming an increasing share of the power grid,” Hersam said. “This is an unsustainable path if we continue to rely on traditional computer hardware.”

The study, “Reconfigurable Mixed-Kernel Heterojunction Transistors for Personal Support Vector Machine Classification,” was supported by the U.S. Department of Energy, the National Science Foundation, and the Army Research Office.