<!channel> 🚀 Unlock the Power of Custom Silicon for Edge ML with chipIgnite ML 💡 Introducing chipIgnite ML – Efabless’ latest custom chip platform, designed specifically for ML edge applications. Experience 10x the performance and power efficiency of traditional solutions, enabling you to build smarter, faster, and more efficient devices for keyword spotting and beyond. Partnering with SensiML, this solution empowers engineers to: • Develop ultra-efficient sensor inference algorithms • Leverage custom silicon tailored to your edge application needs • Utilize open-source tools to reduce development time and cost Together, Efabless and SensiML provide a complete end-to-end solution, eliminating barriers to IoT innovation and giving you the control you need for breakthrough ML edge products. 🔋 10x More Efficient | 🔧 Custom Silicon | 🛠️ Open-Source Tools https://efabless.com/news/press-release-efabless-unveils-new-custom-chip-platform-designed-for-edge-ml-products

This sounds like an exciting development! I'm curious about the customization options available with the chipIgnite ML platform. How customizable is the final chip in terms of internal RAM, the NoC configurations between components, and high-level software integration? What is the extend of flexibility it offers for specific edge ML applications?

That's a great positioning ! sky130 is a great playground to bring-up full custom chips: without EDA tool costs (if you have a design team able to deal with it), + the increasing number of ready to use IPs reaching the marketplace, either free or paid (those long time missing SRAM instances) + the very reasonably priced prototyping infrastructure down to QFN48 chips + no NDA gating the access for first evaluation… I think NDA could bring more IP catalog stuff already silicon proven (example: the exact SRAM 6T single port form factor your need, access to twoPort memory or dualPort memory, possibly more sophisticated analog/mixed signal IPs, maybe some qualified RRAM instances :-) but that could be seen as a refinement