Recent advancements in artificial intelligence (AI) automation are being driven by groundbreaking research in photonics and optoelectronics, offering significant implications for the future of business processing and operational efficiency.
A team comprised of researchers from the Massachusetts Institute of Technology (MIT), Enosemi, and Periplous has developed an innovative fully integrated photonic processor capable of executing all primary computations associated with a deep neural network entirely through optical means. The chip, which is fabricated using commercial foundry processes, features three layers of devices designed to carry out both linear and nonlinear operations.
One of the significant challenges encountered in the project involved the implementation of nonlinear operations on the chip. To address this, the researchers designed unique nonlinear optical function units (NOFUs) that effectively combine electronic and optical elements. The process begins with encoding the parameters of a deep neural network into light, followed by the use of an array of programmable beamsplitters to conduct matrix multiplication on these inputs. Subsequently, the data is processed through the programmable NOFUs, which apply nonlinear functions by extracting a small fraction of light to photodiodes that convert optical signals to electrical current. This methodology reduces energy consumption and negates the need for external amplifiers.
In a statement to Semiconductor Engineering, Saumil Bandyopadhyay, a visiting scientist in MIT's Quantum Photonics and AI Group, remarked, “We stay in the optical domain the whole time, until the end when we want to read out the answer. This enables us to achieve ultra-low latency.” This low latency rendered it possible for a deep neural network to be trained on the chip efficiently. Bandyopadhyay added, “This is especially useful for systems where you are doing in-domain processing of optical signals, like navigation or telecommunications, but also in systems that you want to learn in real-time.” Remarkably, the optical device accomplished key computations for a machine-learning classification task in under half a nanosecond, obtaining an accuracy rate exceeding 92%.
Similarly, researchers at the Tokyo University of Science have fabricated a revolutionary self-powered device leveraging dye-sensitized solar cells for efficient edge AI processing of time-series data. The device utilises squarylium derivative-based dyes, integrating optical input, AI computation, analog output, and power supply functions at the material level. This innovation allows the system to operate with ultra-low power consumption, enabling it to classify human movements, including bending, jumping, running, and walking, with over 90% accuracy. Takashi Ikuno, an associate professor in the Department of Applied Electronics at TUS, referenced the design’s foundational inspiration from the afterimage phenomenon of the eye, stating, “In order to process time-series input optical data with various time scales, it is essential to fabricate devices according to the desired time scale.”
In further developments, a collaborative team from Huazhong University of Science and Technology and Wuhan National Laboratory for Optoelectronics has introduced a large-scale optical programmable logic array (PLA). This advanced array employs parallel spectrum modulation to create an 8-input system, which successfully manages more complex logic operations. The optical PLA has demonstrated the ability to execute advanced functions such as decoders, comparators, adders, and multipliers, illustrating its functionality by running models like Conway’s Game of Life and other cellular automata without the reliance on electronic components for nonlinear computing.
These advancements in photonic and optoelectronic technologies showcase the potential transformation of AI automation within various business applications, emphasising enhanced efficiency, reduced energy consumption, and improved processing speeds as integral components shaping future business practices.
Source: Noah Wire Services