Jim Keller on Open Architectures, Chiplets, and AI Computing

Redefining AI Architecture

• 💡 Jim Keller, CEO of Tenstor, emphasizes that openness, modularity, and simplicity are crucial for the future of AI computing.
• 🔑 His design philosophy, inspired by "The Systems Bible," states that complicated systems cannot be fixed and require re-engineering into simpler, modular components.
• 🚀 The shift from processor-centric to data-centric systems means performance now hinges on efficient data movement across various scales.

Advancing with Chiplet Technology

• 🧩 Tenstor's Open Chiplet Atlas (OCA) provides a framework for modular chiplet-based designs, addressing challenges in test, power management, and negotiation.
• ⚡ Chiplets enable breaking down complexity, balancing yield, and offering flexibility, moving beyond traditional monolithic chip designs.
• ⚠️ Committees are often too slow for rapid innovation in chiplet standards, necessitating a more agile, hands-on approach to development.

The Power of Open Ecosystems

• 🌐 RISC-V is championed as an open CPU architecture platform, fostering innovation and collaboration, especially in academic and emerging markets.
• 🧠 Tenstor's Tensic architecture is being developed as an open AI architecture, allowing for modification and experimentation by the community.
• ✅ An open-source model for hardware, with co-simulation and published collateral, is vital for accelerating development and enabling partners.

Fostering Collaborative Computing

• 🌍 Many regions seek sovereign technology ownership, driving demand for local tech stacks and open platforms.
• 🤝 Tenstor provides a full technology stack (IP, chiplets, software, models) and licenses components, enabling partners to innovate without building everything from scratch.
• 📈 This approach supports fit-for-purpose, cost-effective, and energy-efficient solutions tailored for diverse markets.

Future of Architectural Innovation

• 🌱 New architects should focus on open technologies in smaller companies to maximize impact and gain hands-on experience.
• ⚡ Emphasize velocity and rapid iteration, leveraging tools like AI for code generation, rather than slow, meeting-heavy processes.
• 🎯 The goal is to move from architectural concept to silicon faster, with independent, fast-moving components.