New Analog Chip Architecture Features high precision

Analog electronics is often the forgotten part of electronics. Of course the discipline is very important to the industry as it is how we interface the real world to the digital world. However, in many applications, the first component in systems after the sensor is the analog to digital converter which converts the signal for processing, and the last component in the system is the digital to analog converter, which converts it back to analog to be used by the actuator. Even power is digitally controlled in many cases these days. One area that is only partially affected is audio, where HI-FI purists claim that the conversion to digital and back has a noticeable effect on the sound quality and therefore prize all analog systems.

However, the world is changing. The Internet of Things has led to the installation of sensors almost everywhere, and although a lot of those sensors only take small readings that are easily converted, others, such as video sensors take huge amounts of data at a time. That data often has to be converted, processed, stored, converted back to analog and a response sent very quickly, for example in the case of a vehicle’s image sensing system. That data also has to be processed in real-time along with data from other sensors, such as LIDAR and RADAR. Pushing those large amounts of data around the system can often lead to a slowdown, especially in memory. If the computation could be done in the analog domain, things could be much easier. It should also require less power, which would be a bonus, especially for any application that runs on batteries. Last week in this column, I wrote about one team of researchers who were trying to achieve analog computing using capacitance. Another team think that memristors could be the ideal solution. Wikipedia defines memristors as non-linear two-terminal electrical component relating electric charge and magnetic flux linkage. In short, they are small components that are able to store and processes data very efficiently.

The researchers from University of South Carolina’s Viterbi School of Engineering Electrical and Computer Engineering, led by professor, J. Joshua Yang, had previously published a paper describing how they had found a way to tweak a memristor to achieve higher precision. Now, the lab has enhanced that process to achieve even higher precision with the same memristors. This development could greatly extend the applications of the technology to applications as complex as neural networks. Further, the innovation can be used other types of memory technologies, including magnetic memories and phase change memories.

Normally, it is very challenging to quickly program an analog device precisely to a target value. Yang’s circuit architecture and algorithm to do exactly that, making analog computing much more attractive for many applications.

Yang describes the new technique as having “higher efficiency and higher speed with accuracy of the digital systems”. This type of improvement is critical as it will allow the innovation to be applied to train neural networks for artificial intelligence and machine learning, Currently, that process requires expensive and power hungry digital systems. The innovation will also allow the development of analog computing for new applications, such as scientific computing.