Advanced Computing in the Age of AI | Thursday, February 29, 2024

Untether AI Joins UCIe Consortium to Drive Chiplet Tech and Energy-Centric AI Acceleration 

TORONTO, Nov. 28, 2023 -- Untether AI, a leader in energy-centric AI inference acceleration, today announced that it has joined Universal Chiplet Interconnect Express (UCIe) Consortium, the industry organization that has developed an open specification that defines the interconnect between chiplets within a package. As the leader in energy-centric AI acceleration, Untether AI sees heterogeneous compute with die-to-die interconnect as the next logical step for energy-centric compute architectures.

Compute workloads have radically changed in the past several years with the emergence of AI workloads. The increasing complexity of AI and conventional software algorithms has elevated the demand for heterogeneous computing using domain-specific architectures. This means companies are deploying a combination of AI accelerators, CPUs, GPUs, memory, and networking technologies to create the most efficient hardware architectures for their workloads. This can pose significant challenges in terms of energy consumption and data transport, especially as neural network models explode in size and further exacerbate the I/O energy requirements.

UCIe addresses these critical issues by offering a universal standard for die-to-die interconnects, enabling energy-efficient and seamless communication between different components. The recent release of the UCIe 1.1 specification expands support to automotive applications such as autonomous vehicles, a key vertical for Untether AI.

Driven by this vision of plug-and-play chiplets, Untether AI will enable the development of compact and high-performance compute systems across a number of domains, ranging from high-performance computing to edge applications. Untether AI's speedAI devices have a groundbreaking at-memory compute architecture that seamlessly supports chiplet methodologies due to its unique spatial scalability and I/O ring NOC (network on chip). Furthermore, the spatial architecture of Untether AI's technology enables the creation of chiplets of varying sizes, catering to specific application requirements.

"We are pleased to join the UCIe Consortium to advance the future of AI acceleration," said Arun Iyengar, CEO, Untether AI. “By leveraging chiplet technology and UCIe's universal standards, Untether AI will deliver energy efficient AI acceleration chiplets to the market. We are already seeing great acceptance of UCIe from our customers and are excited to see the proliferation of products with UCIe.”

"We are excited to welcome Untether AI into the UCIe Consortium and look forward to their contribution to the UCIe specification,” said Dr. Debendra Das Sharma, UCIe Consortium chairman. “Untether AI’s contribution will help showcase the benefits of UCIe, a low-power, high-speed, and open industry standard chiplet interface."

About Untether AI

Untether AI provides energy-centric AI inference acceleration from the edge to the cloud, supporting any type of neural network model. With its at-memory compute architecture, Untether AI has solved the data movement bottleneck that costs energy and performance in traditional CPUs and GPUs, resulting in high-performance, low-latency neural network inference acceleration without sacrificing accuracy. Untether AI embodies its technology in runAI and speedAI devices, tsunAImi acceleration cards, and its imAIgine Software Development Kit. Founded in Toronto in 2018, Untether AI is funded by CPPIB, GM Ventures, Intel Capital, Radical Ventures, and Tracker Capital.

Source: Untether AI