The poster and slides from Chaoqun's @chaoqun_L#DATE2025 contribution "Towards Reliable Systems: A Scalable Approach to AXI4 Transaction Monitoring" are now online. Check them out: https://t.co/9ZqgMtY8Si & https://t.co/7d7Ef4b7Hn
We propose a method to accelerate the Softmax function, a key bottleneck in Transformer models, by integrating a custom exponential instruction into Snitch. See "VEXP: A Low-Cost RISC-V ISA Extension for Accelerated Softmax Computation in Transformers" https://t.co/yFVRAqSZ0U
Addressing the softmax bottleneck in @pulp_platform clusters, at a minimum hardware cost. This is critical for efficient attention computation, especially when GEMM/V is accelerated by the redmule tensor engine!
Check out our CF25 paper "CVA6-VMRT: A Modular Approach Towards Time-Predictable Virtual Memory in a 64-bit Application Class RISC-V Processor" a CVA6 extended with HW support for predictability in virtual memory access with minimal area overhead https://t.co/sCIIdL7uVJ
Here is our latest a brief on a 16nm, reliable, time-predictable heterogeneous SoC with multiple programmable accelerators. See now "A Reliable, Time-Predictable Heterogeneous SoC for AI-Enhanced Mixed-Criticality Edge applications" on arXiv: https://t.co/yin28nC4Ra @Ang_93
We present a lightweight, modular, technology-independent & open-source real-time extension to AXI4 interconnect. See "AXI-REALM: Safe, Modular and Lightweight Traffic Monitoring and Regulation for Heterogeneous Mixed-Criticality Systems" https://t.co/1D0qzH6kwn @ThommyThomaso
Next for #DATE2025 in Lyon, @chaoqun_L will present a versatile Transaction Monitoring Unit for AXI4-based SoCs. Find the paper on arXiv under "Towards Reliable Systems: A Scalable Approach to AXI4 Transaction Monitoring" https://t.co/7ZaxpkN2ec @ThommyThomaso@aottaviano96
Just before the end of the year, here is our paper "Open-Source Heterogeneous SoCs for AI: The PULP Platform Experience" with focus on designing heterogeneous AI acceleration SoCs. Find it here: https://t.co/gspYlPcyoG @arunax@Ang__93@duav_red@GiuseppeTag
We present "A Flexible Template for Edge Generative AI with High-Accuracy Accelerated Softmax & GELU" based on a 8-core RISC-V cluster with a 24 × 8-PEs tensor processing unit and a novel accelerator for BFloat16 softmax. See https://t.co/yyEQRP7RUE @YvanTortorella@Ang__93
Tadaa! Carfield has just printed its first hello world string. Thanks to DUTCTL presented at #RISCV summit and the amazing work on standardizing Cheshire, we could bring Carfield up in less than 4 hours! https://t.co/9rUc4miTs3 https://t.co/Nr1CBJx7ty @ThommyThomaso
OSHW24 collocated with CF'24 takes place next week. We will be there with Gigabit, DMA-enhanced Open-Source Ethernet Controller for Mixed-Criticality Systems @chaoqun_L & NARS: Neuromorphic Acceleration through Register-Streaming Extensions on RISC-V Cores https://t.co/q1XH6DmOA0
Nice opportunity to submit a paper! @LucaBeniniZhFe, @duav_red & @Ang__93 are organizing SSH-SoC collocated with #DAC2024. @JBalkind, John East, Jon Michelson @zeroRISCinc, Jose Martins #UMinho & Rafail Psiakis @TIIuae among invited speakers. Learn more: https://t.co/POQIwKhlN6.
Save the dates for CF24-OSHW Workshop on Open-Source Hardware May 7th - 9th, 2024 in Ischia that we are co-organizing together with @openhwgroup, Uni Bologna and Polytechnic of Torino. Read all about it here: https://t.co/ZEanzV1QVt
We are thrilled to announce that Jon Michelson, VP of Engineering and co-founder of zeroRISC, will be delivering a talk at the upcoming second edition of the Safety and Security in Heterogeneous Open System-on-Chip Platforms (SSH-SoC) Workshop.
Our congratulations to Angelo @Ang__93 who just won the ISSCC 2023 Outstanding Forum Presenter Award for his last year's contribution "Is an AI Accelerator All You Need?". Check out the talk https://t.co/RKsWn6hJ0W
We are delighted to release Carfield, our open-research heterogeneous platform for safety, resilient & time-predictable systems. It now comes with exhaustive docs & an end-to-end getting started guide for tests & apps. Give it a try https://t.co/jVdAkKbfRx https://t.co/Jo71XUb2Uj