UCIe™ Consortium member Untether AI® recently shared their thoughts on the importance of UCIe technology in AI applications as well as its contribution to the Consortium and the UCIe specification development.
Can you share a brief introduction of Untether AI?
Founded in 2018, Untether AI aims to extend artificial intelligence beyond data centers to the network edge. Our unique at-memory architecture for neural network inference offers unmatched performance and efficiency. Placing memory next to compute within a spatial layout reduces data movement for each operation, which lowers power consumption and latency. The runAI® and speedAI® product families, showcasing this architecture, lead the industry in inference operations per second per Watt. The scalability of our spatial architecture is well suited to chiplet methodologies, accommodating various sizes and applications.
What prompted Untether AI to join the UCIe Consortium?
With AI and software algorithms growing in complexity, there is a surge in the need for heterogeneous computing, which combines different technologies like CPUs, GPUs and AI accelerators like us. This trend poses challenges in energy use and data transport, exacerbated by growing neural network models. The UCIe specification offers solutions by providing a universal interconnect framework, enhancing communication efficiency among heterogeneous compute components. Our membership in the Consortium supports the development of this crucial technology, particularly beneficial for automotive applications like autonomous vehicles and cloud infrastructure.
What is the importance of a chiplet ecosystem to Untether AI?
Standardized connectivity is vital for addressing the diverse and power-sensitive needs of edge applications. Chiplets, offering modular and efficient solutions, are better suited than traditional monolithic systems on chips for diverse workloads. The architecture offers flexibility to changing demands of AI workloads at the edge, offering the ability to build devices with the right compute for the right job. The UCIe specification, promoting interoperability, reduces development time and cost while accommodating best-in-class technologies, a boon for heterogeneous designs.
Can you share some UCIe technology use cases that Untether AI is bringing to the industry? Are there any specific market segments that will benefit most from UCIe?
In the automotive industry, the shift towards software-defined vehicles and advanced AI applications is accelerating. The UCIe specification’s adaptation for automotive needs aligns with this shift, allowing for the efficient integration of high-performance AI computing resources. As automotive architectures evolve, chiplet technology becomes increasingly crucial.
As AI inference workloads continue to increase in cloud infrastructure due to the explosion of generative AI, chiplets allow the development of devices that can span workloads from high performance compute to large language models.
What would you say to a company considering joining the UCIe Consortium and supporting the chiplet ecosystem?
The chiplet ecosystem, facilitated by the UCIe Consortium, represents a significant advancement in semiconductor industry history. It offers a cost-effective, low-risk pathway for SoC innovation. Broad industry participation is essential for UCIe’s success. We encourage companies to consider joining and contributing to this transformative technology.
Comments