Scientists have developed a neuromorphic chip to drive AI applications that can perform calculations directly in memory without the need for a network connection with the cloud. In addition, the chip consumes a small amount of energy to make it more efficient than other chips. The discovery is expected to enable AI to be used in a range of edge devices where it can host sophisticated tasks without relying on a centralized server.
The NeuRRAM chip has proven to be more efficient than compute-in-memory chips and can produce as accurate results as conventional chips. With this, the chip could have applications in tasks such as image recognition and reconstruction, and voice recognition.
AI computing requires both power and computational capability. Most AI applications on edge devices require data to be moved from the device to the cloud, where it is processed. The data is then moved back to the device. This is because most edge devices are battery-powered and have limited power that they use for computing.
Developed by engineers at the University of California, the NeuroRam chip reduces this power consumption, making edge devices smarter, stronger and more accessible. Moreover, it also enhances data security as there are some data privacy risks involved in transferring data from device to cloud.
The process of transferring data is considered a cumbersome task. “It’s the equivalent of doing an eight-hour commute for a two-hour workday,” Explained Weier Van, a PhD graduate of Stanford University who worked on the chip at UC San Diego. He is also a co-author of the study published in Nature,
The team used a type of non-volatile memory called resistive random-access memory that enables computation within a memory without the need for a separate computer unit. While compute-in-memory is not a new method, the NeuraM chip is different because it offers great efficiency and flexibility for diverse AI applications while maintaining similar accuracy.
The researchers demonstrated the chip’s capability by running various functions on it and saw impressive results that were on par with existing digital chips.
