Scientists develop a novel analog chip structure with exceptional accuracy

Although digital computing dominates, much of the data we encounter, such as images from cameras or temperature and sound readings, is initially analog and must be converted to digital for accuracy. Consider an autonomous vehicle: it must swiftly and accurately process road data to make split-second decisions, requiring rapid, energy-efficient, and precise analog-to-digital conversion.

What if newly developed analog chips could combine the precision of digital computing with the energy-saving and high-speed advantages of analog computing? A study titled "Programming memristor arrays with arbitrarily high precision for analog computing" has been published in the journal Science.

A memristor, a compact circuit component, efficiently stores and processes data. In a previous study led by J. Joshua Yang, a professor at USC Viterbi School of Engineering's Electrical and Computer Engineering department, researchers successfully enhanced a memristor for unparalleled precision.

Yang's lab, part of USC Viterbi and its School of Advanced Computing, focuses on devising computing devices. They have devised a new circuit and architecture to achieve even greater precision using the same memristors. This advancement could significantly broaden the applications of such technology, moving beyond conventional low-precision applications like neural networks.

Furthermore, Yang notes that this advancement extends to other memory technologies, including magnetic memories, which employ the same device as read-heads in magnetic hard disk drives, and phase change memories, which use the same material as compact disks (CDs).

Typically, achieving precise programming of an analog device to a target value swiftly poses significant challenges. However, Yang's lab has devised circuit architecture and corresponding algorithms to accomplish precisely that. This breakthrough renders analog computing using analog devices much more appealing for numerous applications.

According to Yang, this innovation offers "greater efficiency and speed with the accuracy of digital systems."

Such enhancements are crucial, Yang emphasizes, as they can be leveraged to train neural networks necessary for advancing artificial intelligence (AI) and machine learning (ML), which, until now, have been feasible only at considerable expense using digital systems. Moreover, this innovation will facilitate new applications beyond AI and ML, such as scientific computing for tasks like weather forecasting.