Facebook pixel

Sr. Principal SOC Fabric Architect
Confirmed live in the last 24 hours
Toronto, ON, Canada
Experience Level
Desired Skills
Data Analysis
Data Structures & Algorithms
  • Strong grasp of NoC topologies, routing algorithms, queuing, traffic scheduling, and QoS requirements
  • Expertise in cache coherency protocols (AMBA CHI/AXI protocol), DDR/LPDDR/GDDR memory technology, and IO technology (PCIe/CCIX/CXL)
  • Proficient in C/C++ programming. Experience in the development of highly efficient C/C++ CPU models
  • Collaborate with the software team and platform architecture team to understand fabric bandwidth and latency requirements and real-time constraints for AI accelerator, CPU, security, and networking traffic. Devise QoS and ordering rules among the CPU, accelerator, and IO coherent/non-coherent traffics
  • Identify representative traffic patterns for the software applications. Perform data-driven analysis to evaluate fabric topology, QoS, memory architecture , and u-architecture solutions to improve performance, power efficiency, or reduce hardware
  • Create directory-base cache coherency specification to satisfy performance requirements of coherent multiple-cluster CPU system and accelerator. Tradeoff protocol complexity and performance requirements
  • Design cache hierarchy to create best performance
  • Set SoC architecture direction based on the data analysis and work with a cross-functional team to achieve the best hardware/software solutions to meet PPA goals
  • Develop a SoC cycle-accurate performance model includes memory sub-systems, directory-based coherence cache controllers, fabric interconnects, and fabric switches that describe the microarchitecture, use it for evaluation of new features
  • Collaborate with RTL and Physical design engineers to make power, performance, and area trade-offs
  • Drive analysis and correlation of performance feature both pre and post-silicon
Desired Qualifications
  • Prior experience or strong understanding of traffic patterns for ML/AI algorithms in a heterogeneous computation system is a plus
  • Prior experience on formal verification of cache coherence protocols is a plus

51-200 employees

Computer processor architecture manufacturer
Company Overview
Tenstorrent is on a mission to address the rapidly growing compute demands for software 2.0. The company designs processors that are optimized for neural network inference, training and can also execute other types of parallel computation.
Company Core Values
  • Collaboration
  • Curiosity
  • Commitment to solving hard problems