Épisodes

  • Memory Meets Compute: How Hybrid Bonding and 3D Memory Could Transform AI Inference

    Memory Meets Compute: How Hybrid Bonding and 3D Memory Could Transform AI Inference
    Oct 29 2026
    n this episode of Tech Threads: Weaving the Intelligent Future, Baya Systems’ Nandan Nayampally sits down with Charlie Cheng, founder and CEO of TeRAM, to discuss the AI memory wall and the industry’s shift toward alternative memory architectures. With Nvidia licensing Groq’s technology and Cerebras drawing attention to wafer-scale computing, the race to bring memory closer to compute is reshaping how AI systems are built.

    Charlie draws on a career spanning microprocessors, EDA, IP licensing, and ASICs, including his time as CEO of Kilopass, the memory IP licensor acquired by Synopsys. He explains why AI inference changes the memory problem: during decode, every token requires a pass through the entire model, so memory bandwidth now stands alongside capacity as a limiting factor in generating each profitable token. What was once an engineering concern has become a direct cost on the CFO’s P&L.

    The conversation covers why on-die SRAM is no longer scaling, why HBM supply is constrained, and why 3D integration through hybrid bonding is the natural next step for the data center, on-prem AI, and edge devices such as AR/VR glasses. Charlie also shares TeRAM’s goal of delivering 64GB of memory capacity and about 200TB/s of bandwidth close to compute. That could shrink AI models from multiple racks to less than one and cut total power by roughly 7x.

    As memory moves closer to compute, the pressure shifts to how data moves across the chip. Nandan and Charlie discuss how hybrid bonding and chiplets are changing network-on-chip topologies, partitioning, and routing, and why efficient data movement is the other half of breaking the memory wall.

    Tune in for an expert look at how new memory architectures, chiplets, and efficient data movement could help unlock faster, more power-efficient, and more scalable AI inference.

    Recorded and produced by ROC Vox Recording & Production, LLC Rochester, NY
    rocvox.com
    Afficher plus Afficher moins
    32 min
  • Scaling Custom Silicon in China: Technology Is Only the Beginning

    Scaling Custom Silicon in China: Technology Is Only the Beginning
    Sep 24 2026
    In this episode of Tech Threads: Weaving the Intelligent Future, Baya Systems Founder and CEO Sailesh Kumar joins AdoreSys Founder and CEO KengSu Teoh to discuss the companies’ new partnership and what it means for semiconductor customers across China and the broader Asia-Pacific region.
    Hosted by Nandan Nayampally, the conversation explores the rapid growth of custom silicon, the increasing importance of data movement and interconnect technology, and how Baya and AdoreSys are combining complementary strengths to support complex SoC and chiplet designs.

    From Technology Adoption to Implementation
    Sailesh and KengSu discuss how Baya and AdoreSys can combine their complementary strengths to support customers across the design journey, from evaluating advanced interconnect technology to silicon implementation.

    Inside the Conversation
    • Why data movement and connectivity are becoming defining challenges in modern computing systems
    • How the shift toward custom silicon is changing customer requirements and development models
    • Why chiplet architectures create new opportunities and new connectivity challenges
    • What semiconductor design teams in China value when evaluating and adopting technology
    • How local engineering relationships can accelerate technical engagement and customer support
    • What Baya Systems and AdoreSys hope to accomplish through their partnership

    Recorded and produced by ROC Vox Recording & Production, LLC Rochester, NY
    rocvox.com
    Afficher plus Afficher moins
    44 min
  • The Memory Maze: Copper, Optics, and the Future of AI Data Movement

    The Memory Maze: Copper, Optics, and the Future of AI Data Movement
    Aug 18 2026
    As AI models grow larger and more autonomous, every query requires extraordinary volumes of data to move quickly between memory and compute, making data movement one of the defining challenges for next-generation AI infrastructure.
    In this episode of Tech Threads, Nandan Nayampally speaks with Srujan Linga, CEO and co-founder of Kandou AI, about why the industry’s so-called “memory wall” is actually a multidimensional “memory maze,” shaped by different requirements for bandwidth, latency and capacity.
    They explore the limitations of relying exclusively on high-bandwidth memory, the untapped potential of copper interconnects, and how technologies such as Chord Signaling and Copper MIMO can increase bandwidth while reducing power and extending reach. The conversation also examines where copper and optics can complement one another, how advanced interconnects could turn an entire PCB into a larger multi-chip system, and why the economics and scalability of AI infrastructure must be considered alongside raw performance.
    The result is a system-level perspective on what it will take to build more efficient, scalable and accessible AI infrastructure.
    Afficher plus Afficher moins
    44 min
  • Chiplets, RISC-V, and Europe's AI Silicon Ambition

    Chiplets, RISC-V, and Europe's AI Silicon Ambition
    Jul 28 2026
    Europe has the technical talent and research base to compete in advanced semiconductor design. Turning that into globally competitive AI and HPC systems takes more than investment. It needs scalable architectures, open ecosystems, and real collaboration across the supply chain.
    In this episode of Tech Threads, Baya’s CCO, Nandan Nayampally talks with Cesc Guim, CEO of Openchip, about building next-generation processors and computing platforms in Europe, why chiplet-based design is becoming central to scalable, specialized architectures, and how RISC-V opens up flexibility that fixed-ISA approaches can't match.
    They also dig into the hardest problem in modern system design: moving data efficiently between processors, accelerators, memory, and chiplets. At scale, performance depends less on adding compute and more on getting the interconnect right: bandwidth, latency, and efficiency for demanding AI workloads.
    Cesc also shares his take on the challenges and opportunities and the thrill of building a semiconductor company in Europe, forming the right technology partnerships, and turning regional ambition into products that compete globally.
    In this episode:
    • How Openchip is approaching AI and HPC system design
    • Why chiplets are becoming central to scalable, specialized architectures
    • Where RISC-V fits in enabling open, differentiated designs
    • Why interconnect performance is the bottleneck for AI system efficiency
    • How partnerships, especially with critical pieces like Baya’s IP, cut complexity, risk, and time-to-market
    • What Europe needs to convert semiconductor investment into competitive products
    Afficher plus Afficher moins
    44 min
  • Beyond the CPU vs GPU War: Rethinking AI Compute at the System Level

    Beyond the CPU vs GPU War: Rethinking AI Compute at the System Level
    Apr 28 2026
    In this episode of Tech Threads, Nandan Nayampally, Baya Systems CCO, sits down with Ian Ferguson, Vice President of Vertical Markets and Business Development at SiFive, to unpack one of the most important shifts happening in modern computing: AI is no longer just about scaling compute, it’s about orchestrating complexity.

    As architectures fragment across accelerators, chiplets, and custom silicon, the real challenge is no longer building faster chips. it’s turning all of these elements into a cohesive, high-performance system.

    This conversation explores why the industry is moving beyond the traditional “CPU vs GPU” narrative and toward a system-level approach where performance is defined by how effectively compute, memory, interconnect and software work together.

    From the growing momentum behind RISC-V to the rise of heterogeneous compute environments, the discussion highlights a clear trend: the future won’t be defined by a single dominant architecture, but by optimized combinations of technologies tailored to specific workloads.

    That shift introduces a new layer of complexity.

    Key themes explored in this episode include:
    - Why data movement is emerging as the primary constraint in AI systems
    - How efficiency metrics like “tokens per dollar” are reshaping design priorities
    - The shift toward purpose-built architectures across data center, automotive, and edge applications
    - The role of open ecosystems and interoperability in accelerating innovation
    - Why competitive advantage is shifting from individual components to full system design

    If you’re interested in where AI is headed, this is a must-watch conversation on the forces shaping the future of compute and what it takes to stay ahead.
    Afficher plus Afficher moins
    49 min
  • Inside the AI Bottleneck: Data Movement, Chiplets, and System Scaling

    Inside the AI Bottleneck: Data Movement, Chiplets, and System Scaling
    Mar 27 2026
    For the last decade AI has been driven by one thing, more compute: bigger models, more accelerators, higher throughput.

    But as NVIDIA’s Jensen Huang recently highlighted at GTC, the industry is hitting a different kind of wall, one that hasn’t received nearly as much attention.

    The real constraint is no longer just compute. It’s data movement.

    To its credit, Nvidia has pushed this frontier with innovations like NVLink Fusion, and continued investment in connectivity AI dataflow architectures. But the challenge is bigger than any one company.

    As AI systems scale to hundreds - and even thousands - of processors, performance is increasingly defined by the ability to efficiently move, synchronize, and manage data across increasingly distributed architectures that can orchestrate data across chiplets, nodes, and entire racks.

    In this episode of Tech Threads, we bring together a panel of deeply experienced technologists, architects and leaders from companies like Intel, Arm, Altera, Texas Instruments, and Arteris - individuals who have helped shape modern compute, interconnect standards, and system architecture.

    Together they explore what is really changing beneath the surface: why traditional scaling approaches are breaking down, how coherent interconnects and network-on-chip architectures are evolving, and why system-level thinking is becoming essential.

    They also dive into the growing complexity introduced by chiplet-based designs, heterogeneous compute, and distributed memory systems and what it takes to maintain performance, efficiency, and programmability at scale.

    This is not just a technology shift, it’s an architectural reset.

    If you’re building or thinking about next-generation AI systems, this conversation gets to the heart of what matters next.
    Afficher plus Afficher moins
    54 min
  • From Arduino to AI Infrastructure: Scaling the Next Wave of Computingl

    From Arduino to AI Infrastructure: Scaling the Next Wave of Computingl
    Jan 21 2026
    What do Arduino, IoT, edge AI, and Nvidia-era data centers have in common? They all depend on ecosystems: people, platforms, and momentum.
    In this episode, Sander Arts joins Baya’s Chief Commercial Officer and Tech Threads host Nandan Nayampally for a wide-ranging, candid conversation on how breakthrough technology actually scales.
    Sander brings a rare operator’s perspective shaped by 25+ years scaling global technology companies across semiconductors, enterprise software, and AI. As the founder of Orange Tulip Consultancy, he serves as a Fractional CMO and growth advisor, helping leadership teams turn deep technology into real-world adoption.
    Together, Nandan and Sander explore how communities, developer access, and platform ecosystems turn deep technology into real-world adoption, and why timing and openness can be just as critical as technical performance.
    The discussion moves from the maker-era lessons of Arduino and IoT to today’s AI infrastructure boom, unpacking why scaling “long-tail” customers is both an opportunity and an operational challenge, and how edge AI and data center markets are evolving in parallel. They also debate the art of “opening the kimono,” how standardization and middleware shape adoption, and why capital intensity and speed often determine whether innovation stays local or becomes global.
    They close by looking ahead at emerging trends like robotics, neo-cloud architectures, quantum with real customers, and the networking backbone powering AI’s future, and how these shifts intersect with Baya’s view of increasingly complex, software-driven systems.
    Afficher plus Afficher moins
    43 min
  • The Architecture of "Open" Intelligence

    The Architecture of "Open" Intelligence
    Oct 14 2025
    In this episode of TechThreads: Weaving the Intelligent Future, legendary chip architect Jim Keller joins Nandan Nayampally, Baya Systems’ Chief Commercial Officer, to explore how openness, modularity, and simplicity are redefining the architecture of intelligence.

    From his early work on Apple’s A4 through A7 processors to today’s AI-driven computing revolution, Jim shares how every leap in performance has come from breaking complexity down into composable, modular layers. Referencing The Systems Bible, he explains why “you can’t fix broken complicated systems”, and why the only path forward is to design simpler components that can scale and evolve together.

    The conversation spans:
    - The AI paradigm shift. Why traditional compute models no longer scale.
    - How data movement, not just compute, has become the new frontier.
    - The rise of chiplets and software-driven fabrics for scalable design.
    - The power of open ecosystems like RISC-V and OCA to democratize AI innovation.
    - Building a path toward sovereign and collaborative compute platforms worldwide.

    Listen as Jim Keller unpacks the engineering philosophy behind building open, intelligent systems and what it means for the future of AI and computing at scale.
    Afficher plus Afficher moins
    44 min