Since cognition operations need to deal with a very large amount of data, how to efficiently process these data becomes very important. The biggest challenge in dealing with large amounts of data is that today's computer architectures are all run by von Neumann architecture, but with the von Neumann architecture. When performing AI applications, a large amount of data needs to be transferred between the processor and the memory, and the transmission of data consumes a lot of resources, resulting in poor performance.
The new architecture proposed by IBM is based on the original von Neumann architecture. In addition to the computational memory unit, the computational memory unit is responsible for a large number of computational tasks. The von Neumann machine is responsible for repeatedly improving the accuracy. Therefore, the system has high accuracy. Rate and high efficiency two advantages.
Among the memory devices of this architecture, IBM uses phase-change memory (PCM). With the fast read/write speed of PCM, it can efficiently process large amounts of data without transferring data to the CPU. The speed of implementation will increase dramatically, and it will also reduce resource consumption.
According to IBM, such an architecture can be implemented through computational memory without sacrificing overall accuracy and overcoming the current energy challenge of the von Neumann architecture.