Date: Jun 24, 2025
Category: Innovation
Recent innovations in AI System-on-Chip (SoC) design have predominantly targeted the acceleration of tensor computations. However, the crucial aspect of tensor manipulation—characterized by large-scale data movement with minimal computation—has received comparatively little attention. Addressing this gap, the Tensor Manipulation Unit (TMU) introduces a groundbreaking architecture that is both reconfigurable and positioned near memory, enabling high-throughput processing of tensor data. By optimizing data movement and manipulation directly adjacent to memory, the TMU significantly reduces latency and power consumption, paving the way for more efficient AI workloads. This advancement not only enhances the overall performance of AI systems but also opens new possibilities for scalable and flexible AI hardware solutions. Explore how the TMU is set to redefine the landscape of AI hardware by bridging the gap between computation and data movement. Read the source »
Share on:
Titan and Yanko Design have launched the "Retrofuturism Challenge," inviting designers to participate in the YD x Titan Design Competition. The challenge offers internships and prizes exceeding...
Qualcomm has introduced the Snapdragon 8s Gen 3 mobile platform, designed to elevate generative AI, photography, and gaming experiences. This cutting-edge technology promises to revolutionize mobile...
Aston University researchers have developed a revolutionary method to transmit data at speeds 4.5 million times faster than typical broadband, potentially revolutionizing internet connectivity.
According to Bloomberg's Mark Gurman, Apple is working on its own on-device large language model (LLM) to power upcoming generative AI features.
Researchers at the University of Tokyo have unveiled Musashi, a humanoid robot capable of driving standard vehicles, pushing the boundaries of autonomous transportation.
Meta's Llama 3.1, the largest open-source AI model, was released on Tuesday. Just a day later, Mistral unveiled Large 2, claiming it rivals the latest models from OpenAI and Meta. Explore the...