Hardware Podcasts

Choosing the Right Technologies for Hybrid AI Chips

EE Times Current · 5 Apr 2024 · 52m

What this episode covers

Dr. Amirreza Yousefzadeh, formerly of imec and Gray Matter Labs, now at the University of Twente, explains the design of the Seneca hybrid neuromorphic chip, which combines conventional digital processing with event-driven neural network architectures. He details the decision to use time-multiplexed neurons and a dual-controller 'big.LITTLE' structure to balance flexibility and power efficiency, aiming to compete with conventional AI accelerators like NVIDIA GPUs. The episode also features discussion from Dr. Giulia D’Angelo of Fortiss and Professor Ralph Etienne-Cummings of Johns Hopkins University on the commercial viability and research directions of neuromorphic engineering.

Most relevant to

early-stage companies

Topics

Mentioned

Companies
imec, Gray Matter Labs, Fortiss, IBM, NVIDIA, Intel, BrainChip
Products
Seneca, TrueNorth, Loihi, ResNet-50
Also
University of Twente, Johns Hopkins University, University College London, Synopsys, Cadence

Related episodes from other shows

Other shows on the same ground, closest first.

  • The Circuit · 48m

    EP 180: HPE Reflections on AI Compute Racks, Analog Analog Analog Semis.

    Ben Bajarin and Jay Goldberg discuss findings from HPE Discover and Computex, focusing on how Dell and HPE are differentiating their AI infrastructure through custom rack engineering, hybrid cooling, and specialized networking strategies. They analyze the shift toward on-premises 'private AI factories' for enterprises and the growing importance of analog components and passive parts like MLCCs driven by the power demands of NVIDIA's Vera Rubin GPUs.

    Mechanical engineeringAnalog & power electronics
    9 chapters

  • The Circuit · 33m

    Episode 23: Emerging Competitive Dynamics in Silicon Thanks AI

    Ben Bajarin and Jay Goldberg discuss the emerging competitive dynamics in the semiconductor industry driven by AI, debating whether AI spending is additive to data center capital expenditure or cannibalizes traditional cloud processor budgets. They analyze AMD's recent AI and Datacenter event, highlighting the MI300X chip and its modular chiplet architecture as a strategic response to NVIDIA's dominance in training workloads. The hosts also examine the shift toward Total Cost of Ownership (TCO) and performance-per-watt metrics in the 'cloud native' segment, where AMD and Intel are competing with ARM-based designs like those from Ampere and Amazon Graviton.

    Chips, FPGAs & semiconductorsPricing, margins & unit economics
    6 chapters

  • The Next Byte Podcast · 28m

    145. The Prius of Artificial Intelligence: Hybrid AI Models

    Hosts Daniel and Forbode discuss Qualcomm's proposal for 'hybrid AI,' a model that splits generative AI workloads between edge devices and the cloud to reduce cost, latency, and energy use. They explain how techniques like quantization and pruning allow large models to run locally on hardware like Snapdragon chips, citing examples in autonomous vehicles and mixed reality. The hosts also analyze the implications for hardware design, noting that rapid AI iteration may force faster hardware refresh cycles for consumer devices like iPhones.

    AI & machine learningMicrocontrollers & dev platforms
    7 chapters

  • Oxide and Friends · 1h 54m

    Heterogeneous Computing with Raja Koduri

    Raja Koduri, former chief architect at Intel and AMD, joins Bryan Cantrill and Adam Leventhal to explain the differences between CPUs, GPUs, FPGAs, and ASICs through historical and analogical lenses. Koduri traces the evolution from 3dfx and Direct3D to CUDA, arguing that NVIDIA’s success stems from a tight coupling of programming and execution models, which creates a sticky ecosystem. He uses transportation analogies to categorize these technologies—cars for CPUs, buses for GPUs, and construction vehicles for FPGAs—and discusses how advanced packaging and memory bandwidth are becoming the primary constraints in modern AI hardware.

    Chips, FPGAs & semiconductorsMicrocontrollers & dev platforms
    8 chapters

  • The Circuit · 51m

    Episode 70!!! NPUs!!!!!!!!

    Ben Bajarin and Jay Goldberg host Austin Lyons and Paul Karazuba to discuss the architecture, market, and future of Neural Processing Units (NPUs). The group debunks the metric 'TOPS per watt' as meaningless, arguing that real-world performance depends on specific network workloads and latency. They explore how chipmakers like Qualcomm, Intel, and Apple are designing NPUs, often by licensing IP or evolving DSPs, and predict that NPUs will become ubiquitous in devices from smartphones to microcontrollers within three to four years.

    Microcontrollers & dev platformsChips, FPGAs & semiconductors
    8 chapters

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by EE Times Current. The episode itself belongs to its publisher.