New AI Tool Predicts the Failure of Electronic Devices

Engineers from the University of Colorado Boulder (CU Boulder) are advancing the process of integrating artificial intelligence with sophisticated computer simulations in an attempt to predict the failure of electronics, such as the transistors used in cell phones.

Published recently in the npj Computational Materials journal, the study was headed by physicist and aerospace engineer, Sanghamitra Neogi.

In their new study, Neogi and her collaborators plotted the physics of small building blocks composed of atoms and subsequently applied machine learning methods to predict the behavior of larger structures produced from those small building blocks. It is kind of like looking at a solo Lego brick to predict the strength of a relatively larger castle.

We’re trying to understand the physics of devices with billions of atoms.

Sanghamitra Neogi, Assistant Professor in Ann and H. J. Smead Department of Aerospace Engineering Sciences, University of Colorado Boulder 

This quest could benefit the electronics that underpin our day-to-day lives, from electric cars and smartphones to emerging quantum computers. According to Neogi, engineers could someday use the researchers’ techniques to pinpoint the weak points in the design of electric components beforehand.

The study is a part of Neogi’s broader view of how the realm of very small things, like the wiggling of atoms, can help build new computers that are more efficient or even computers that are inspired by human brains. The study was co-authored by Artem Pimachev, a research associate in aerospace engineering at CU Boulder.

Rather than wait for years to figure out why devices fail, our methods can give us a priori knowledge on how a device is going to work before we even build it.

Sanghamitra Neogi, Assistant Professor in Ann and H. J. Smead Department of Aerospace Engineering Sciences, University of Colorado Boulder 

Heating Up

Neogi’s latest study targets a big weak point in the electronics sector, that is, hotspots, but this does not mean mobile WiFi hookups.

Neogi elaborated that a majority of the latest computing tools have many imperfections — small flaws in electronic components can cause the heat to accumulate at specific sites, similar to a bicycle slowing down when it goes over rough terrain. These “hotspots” can also make smartphones much less efficient.

According to Neogi, the issue is that engineers inspired by computer models, or simulations, find it difficult to predict the appearance of those weak points beforehand.

We can use physics models to understand systems with approximately 100 atoms in them. But that doesn’t compare to the billions of atoms in these devices.

Sanghamitra Neogi, Assistant Professor in Ann and H. J. Smead Department of Aerospace Engineering Sciences, University of Colorado Boulder 

Neogi believes that machine intelligence can help scientists develop more improved electronics.

From Atoms to Devices

Imagine those individual Lego bricks, which, in this example, are clumps of 16 germanium and silicon atoms — the key ingredients in several computer components.

In the latest analysis, Neogi and her collaborators have designed a computer model that utilizes artificial intelligence to learn the physical characteristics inside those building blocks — or how electrons and atoms combine to determine the energy landscape inside a material. Subsequently, the model can extrapolate from those fundamental blocks to predict energy distribution in relatively larger pieces of atoms.

“It collects information from each individual unit and combines them to predict the final properties of the collective system, which can be made up of two, three or more units,” added Neogi.

Neogi’s research still has a long way to go before they can determine all the possible weak points in a device, similar to the size of a phone. However, up until now, the new model has shown to be effective. Neogi and her collaborators have used this tool to precisely predict the characteristics of many real-world materials made from germanium and silicon.

Neogi is also drawing on her knowledge to learn how heat and energy flow at very small scales to not only enhance present-day devices but also help produce futuristic devices.

In 2019, Neogi was involved in a $1.7 million national effort to study the potential for “neuromorphic” computers, or devices that store and inspect data by emulating the activity of the brain’s neurons.

“What I want to do is poke at this world of atoms in your handheld device and understand how materials and electronics come together to make a device work,” concluded Neogi.

Journal Reference:

Neogi, S & Pimachev, A K (2021) First-principles prediction of electronic transport in fabricated semiconductor heterostructures via physics-aware machine learning. npj Computational Materials. doi.org/10.1038/s41524-021-00562-0.