New Analog Chip Design

From images in cameras, to temperature or sound, the data around us is captured via analog sensors. But most modern computing happens in the digital realm and these inputs must be converted to digital form. What if there was a highly precise, efficient, and low energy way to do so?
We’ll explore in the U.S. National Science Foundation’s “Discovery Files”.
Digital technology is advancing the world around us, but often meets with real world limitations. Autonomous vehicles, for example, need to capture what’s happening around them and make decisions instantaneously. This data, be it road conditions, other vehicles, or people, needs to be converted very quickly with low energy use and high precision results.
NSF-Supported computer engineers at the University of Southern California have designed a new memristor circuit and architecture which could extend applications beyond the traditional low-precision territory, such as neural networks, with higher efficiency and higher speed while retaining the accuracy of digital systems.
This new memristor based system processes data exactly where they are stored, without the need of moving them around. Enabling low-precision analog devices to perform high-precision computing.
The innovative approach can be used for analog computing with other types of analog memory technologies such as phase change or magnetic memory devices.
This circuit architecture will enable new applications not only in artificial intelligence and machine learning, but also in areas such as scientific computing for weather forecasting. This technology is more real than just university research as it has been verified on integrated chips made by a startup company, Tetramem Inc. in Silicon Valley.
To hear more science and engineering news, including the researchers making it, subscribe to "NSF's Discovery Files" podcast.