The data transmitting from/to anywhere has a significant delay and the processing may slow down with the present electrical current. This may lead to our favorite streaming applications would show events with a lag and other critical applications like self-driving vehicles and health applications with a deadly delay. With photonic integrated circuits, the IBM research team found a solution for this latency problem.
IBM Research team has developed a way to reduce latency by using photonic integrated circuits that use light instead of electricity for computing. They also present their photonic processing combination as a non-von Neumann, in-memory computing model. That model performs computations with unprecedented, ultra-low latency and computes density. Their model performs computations at even higher speeds, performing key computational primitives associated with AI models, such as deep neural networks for computer vision in a great efficient manner than ever before.
What is Photonic Integrated Circuit (PIC)?
The photonic integrated circuit (also known as a PIC) is a complex integrated circuit that incorporates a large number of optical devices to form a single photonic circuit. The main difference between a PIC and an electronic IC is that the PIC is analogous to an electronic integrated circuit. Many optical devices such as optical amplifiers, multiplexers, demultiplexers, optical lasers, attenuators, and also detectors are integrated into a photonic integrated circuit. For a large-scale operation of such a device, thousands of optical devices will be integrated into the device.
What is Light and In-Memory Computing (IMC)?
The combination of photonic and IMC can solve the problem of lag so that computation in photonic memory plays an important role in critical and important artificial intelligence programs. With computer memory, the photonic process overcomes seemingly obstacles that are too great to be overcome to the bandwidth of artificial intelligence computing systems based on electronic processors.
There is a physical limitation for the electronic processors, as the number of GPUs you can store in a computer or in a self-driving car is not endless. This problem has led developers to turn to photonic for time-critical apps. It incorporates a photonic processor that has a much higher data modulation speed than that of electronics and can run multiple operations at the same time in the same physical core using "wavelength division multiplexing" (WDM). (This is a technology that multiplies numbers of optical carrier signals on a single optical fiber using different wavelengths of laser lights.) In this way, it provides an additional measure by using the frequency area. In essence, they are able to read using different wavelengths or colors, at the same time.
Findings of the Research
They used a measure called TOPS to assess the number of Operations Per Second, in Trillions, that a chip is able to process. In proof of concept, they obtained experimental data with matrices up to 9 × 4 with a maximum input vector of four at a time, they used non-volatile photonic memory devices based on phase-change memory to store the convolution cores on a chip and used photonic chip-based frequency combs to feed the multi-frequency encoded input. Even with the small 9 × 4 matrices, by employing four multiplexed input vectors and a modulation rate of 14GHz, they obtained an enormous processing speed of two trillion MAC operations (multiply-accumulate) per second or 2 TOPS. The result is impressive since the matrix is so small - although they are not doing many operations, they are doing them so fast that the TOPS figure is still very large.
Finally, Abu Sebastian, Distinguished Research Staff Member, IBM Research who involved in this research said,
Our work shows the enormous potential for photonic processing to accelerate certain types of computations such as convolutions. The challenge going forward is to string together these computational primitives and still achieve substantial end-to-end system-level performance. This is what we’re focused on now.
https://www.ibm.com/blogs/research/2021/01/latency-ai-photonics/
Post a Comment