The most advanced analog AI system.
IBM announced an innovative analog AI processor that can perform difficult deep neural network (DNN) calculations with incredible efficiency and accuracy. The development, detailed in a new study published in Nature Electronics, represents a major step forward in the quest for high-performance AI computation that significantly reduces energy consumption.
Running deep neural networks on common digital computer platforms limits their performance and energy efficiency. These digital systems require continuous data transmission between memory and processor units, slowing down calculations and limiting opportunities for energy optimization. IBM Research tackled these questions using analog AI, which is based on the same principles found in the biological brain. This method uses nanoscale resistive memory devices, specifically phase change memory (PCM), to track synaptic weights.
Read the latest blog: AI and the cloud – the perfect combination
Chip efficacy
By applying electrical pulses to the PCM device, a continuous range of synaptic weight values can be created. This analog approach reduces the need for unnecessary data transfers by performing calculations directly in memory, leading to increased efficiency. The latest CPUs include 64 analog in-memory compute cores, making them the most advanced analog AI systems.
To smoothly switch between analog and digital domains, each core is equipped with a crossbar array of synaptic unit cells and a small analog-to-digital converter. Nonlinear neural activation functions and scaling procedures are also managed by digital processing units within each core. The chip also provides a worldwide digital processing unit and digital communication path for interconnection. To prove the chip’s effectiveness, the research team achieved an impressive accuracy of 92.81% on his CIFAR-10 image dataset. This is extremely high accuracy for an analog AI device.
Computing efficiency compared to previous in-memory computing chips was highlighted by throughput per area, measured in giga operations per second (GOPS) per area. This breakthrough chip represents a major advance in artificial intelligence technology with its energy-efficient architecture and improved performance. The innovative design and amazing capabilities of analog AI chips pave the way to a future where high-speed, low-energy AI computing can be used in a variety of situations. The IBM Research discovery is a watershed event that will drive the development of AI-driven technologies for years to come.
Read: AI and machine learning will change business forever
[To share your insights with us, please write to [email protected]]