This paper is an update of the survey of AI accelerators and processors from past four years, which is now called the Lincoln AI Computing Survey - LAICS (pronounced "lace"). As in past years, this paper collects and summarizes the current commercial accelerators that have been publicly announced with peak performance and peak power consumption numbers. The performance and power values are plotted on a scatter graph, and a number of dimensions and observations from the trends on this plot are again discussed and analyzed. Market segments are highlighted on the scatter plot, and zoomed plots of each segment are also included. Finally, a brief description of each of the new accelerators that have been added in the survey this year is included.

Neuromorphic computing company GrAI Matter has $1 million in pre–orders for its GrAI VIP chip, the company told EE Times. The startup has engagement to date from companies across consumer Tier-1s, module makers (including ADLink, Framos, and ERM), U.S. and French government research, automotive Tier-1s and system integrators, white box suppliers, and distributors. As with previous generations of the company's Neuron Flow core, the company's approach for its GrAI VIP chip uses concepts from event–based sensing and sparsity to process image data efficiently. This means using a stateful neuron design (one that remembers the past) to process only information that has changed between one frame of a video and the next, which helps avoid processing unchanged parts of the frames over and over again. Combine this with a near–memory compute/dataflow architecture and the result is low–latency, low–power, real–time computer vision.

Cerebras has shown off the capabilities of its second–generation wafer–scale engine, announcing it has set the record for the largest AI model ever trained on a single device. For the first time, a natural language processing network with 20 billion parameters, GPT–NeoX 20B, was trained on a single device. A new type of neural network, the transformer, is taking over. Today, transformers are mainly used for natural language processing (NLP) where their attention mechanism can help spot the relationship between words in a sentence, but they are spreading to other AI applications, including vision. The bigger a transformer is, the more accurate it is.

Half a billion years ago something remarkable occurred: an astonishing, sudden increase in new species of organisms. Paleontologists call it the Cambrian Explosion, and many of the animals on the planet today trace their lineage back to this event. A similar thing is happening in processors for embedded vision and artificial intelligence (AI) today, and nowhere will that be more evident than at the Embedded Vision Summit, which will be an in–person event held in Santa Clara, California, from May 16–19. The Summit focuses on practical know–how for product creators incorporating AI and vision in their products. These products demand AI processors that balance conflicting needs for high performance, low power, and cost sensitivity.

Researchers at MIT are employing machine learning techniques to better understand turbulent plasma phenomena in fusion devices. According to MIT News, a new deep learning framework was developed that leverages artificial neural networks to represent a reduced turbulence theory. The research is described in two papers, published in Physical Review E and Physics of Plasmas. If researchers hope to control fusion for energy production, they need a better understanding of the turbulent motion of ions and electrons in plasmas moving through fusion reactors. The field lines of toroidal structures known as tokamaks force the plasma particles; the intent is to confine them long enough to produce significant net energy gains, but that's a challenge with extraordinarily high temperatures but also small spaces.

Edge AI chip startup Deep Vision has raised $35 million in a series B round of funding led by Tiger Global, joined by existing investors Exfinity Venture Partners, Silicon Motion and Western Digital. The company began shipping its first-generation chip last year. ARA-1 is designed for power-efficient, low-latency edge AI processing in applications like smart retail, smart city and robotics. While the company's name suggests a focus on convolutional neural networks, ARA-1 can also accelerate natural language processing with support for complex networks such as long short-term memory (LSTMs) and recurrent neural networks (RNNs). A second-generation chip, ARA-2 with additional features for accelerating LSTMs and RNNs will launch next year.

The ongoing recalibration of AI research and development underscores a fundamental tenet of machine learning: We must learn to crawl before we can walk. Thus far, AI hype has mostly talked the talk rather than walking the walk. Returning to what appear to be engineering first principles, U.S. research efforts are attempting to move beyond current "brittle" AI models that excel at only specific tasks. The goal is developing more generalized models that can adapt much like humans do in new situations. Among those efforts is a Machine Common Sense program overseen by the Defense Advanced Research Projects Agency (DARPA) that seeks to imbue machine learning models with the kinds of commonplace reasoning displayed by among the fastest learners on the planet: infants.

Advanced AI development today is still deeply rooted in 1950s computer science philosophies, including the phrase "garbage in, garbage out." The adage reminds us that an AI model is only as good as the data it's trained on. For everything from advanced cancer screenings to suggesting a new movie, data scientists need large and diverse datasets to train AI models. This can be a significant challenge with real-world data. Often protected for privacy reasons, authentic data can be hard to come by and can also be expensive to source, and potentially not as diverse as desired.

While AI originally was targeted for data centers and in the cloud, it has been moving rapidly towards the edge of the network where it is needed to make fast and critical decisions locally and closer to the end user. Sure, training can be still done in the cloud, but in applications such as autonomous driving, it is important that the time-sensitive decision making (spotting a car or pedestrian) is done closer to the end user (the driver). After all, edge systems can make decisions on images coming in at up to 60 frames per second, enabling quick actions. These systems are made possible through edge inference accelerators that have emerged to replace CPUs, GPUs and FPGAs at much higher throughput/$ and throughput/Watt. The ability to do AI inferencing closer to the end user is opening up a whole new world of markets and applications.

Embedded vision technologies are giving machines the power of sight, but today's systems still fall short of understanding all the nuances of an image. An approach used for natural language processing could address that. Attention-based neural networks, particularly transformer networks, have revolutionized natural language processing (NLP), giving machines a better understanding of language than ever before. This technique, which is designed to mimic cognitive processes by giving an artificial neural network an idea of history or context, has produced much more sophisticated AI agents than older approaches that also employ memory, such as long short-term memory (LSTM) and recurrent neural networks (RNNs). NLP now has a deeper level of understanding of the questions or prompts it is fed and can create long pieces of text in response that are often indistinguishable from what a human might write.