NOC Design Engineer

L&T Semiconductor Technologies; Bengaluru; Vor Ort; INR 2.500.000 - 5.500.000; Vollzeit

Zusammenfassung

L&T Semiconductor Technologies is seeking two NoC Design Leads to own the stacked Network-on-Chip that forms the data backbone of xPU — 4 physical planes delivering 12 TB/s cross-section bandwidth. Architect and implement coherency, routing, QoS, and bandwidth allocation across BFOS arrays, SRAM banks, HBM controllers, UCIe links, and host interface.

This role centers on sustaining extreme bandwidth and latency targets, requiring deep NoC expertise, RTL proficiency, and cross-team collaboration

Qualifikationen

• 10+ years designing on-chip interconnects/NoCs for high-performance SoCs.
• Deep understanding of NoC topologies, routing, QoS, and flow control.
• Strong grasp of cache-coherency protocols (MESI/MOESI-class).
• RTL design in SystemVerilog for routers, arbiters and buffers.

Aufgaben

• Architect and own the 4-plane stacked NoC architecture and bandwidth management.
• Define topology, routing, virtual channels, and arbitration with deadlock-free guarantees.
• Design QoS and bandwidth-allocation mechanisms for multiple memory banks and links.
• Own the coherency protocol and its interaction with SRAM, HBM, and host-interface.
• Drive micro-architecture, RTL, and integration of routers, NIs, buffers, and CDCs.
• Model latency, bandwidth, and congestion with cycle-accurate simulations and traffic models.
• Collaborate across teams on floorplanning and verification coverage.
• Define QoR/performance metrics, mentor engineers, and represent NoC in reviews.

Kenntnisse

On-chip interconnects

NoC topologies

Cache coherency

RTL SystemVerilog

Performance modeling

NoC generators/IP

HBM/UCIe familiarity

Scripting (Python/Perl)

Micro-architecture and docu.

Technical leadership

Tools

SystemVerilog

Jobbeschreibung

Role Summary

We are hiring two NoC Design Leads to own the stacked Network-on-Chip that forms the data backbone of xPU — 4 physical planes delivering 12 TB/s cross-section bandwidth. You will architect and implement coherency, routing, QoS, and bandwidth allocation across the BFOS arrays, banked SRAM, HBM controllers, UCIe links, and the host interface. This role is central to sustaining the extreme bandwidth and latency targets the compute architecture demands.

Key Responsibilities

• Architect and implement the 4-plane stacked NoC delivering 12 TB/s cross-section bandwidth across all on-chip and off-chip endpoints.
• Define topology, routing algorithms, virtual channels, and arbitration to guarantee deadlock-free, high-throughput data movement.
• Design QoS and bandwidth-allocation mechanisms balancing traffic between dual BFOS arrays, 512 SRAM banks, HBM controllers, UCIe, and host.
• Own the on-chip coherency protocol and its interaction with SRAM, HBM, and the host-interface coherency layer.
• Drive micro-architecture, RTL, and integration of routers, network interfaces, buffers, and clock-domain crossings.
• Model and analyze latency, bandwidth, and congestion via cycle-accurate simulation and traffic generators; close performance targets.
• Collaborate with PD on floorplan-aware NoC partitioning across the 4 physical planes and with DFT/verification on coverage.
• Define QoR/performance metrics, mentor engineers, and represent NoC in architecture and integration reviews.

Required Skills & Experience

• 10+ years designing on-chip interconnects/NoCs or coherent fabrics for high-performance SoCs.
• Deep understanding of NoC topologies (mesh, torus, crossbar, stacked/multi-plane), routing, flow control, and QoS.
• Strong grasp of cache-coherency protocols (MESI/MOESI-class, directory-based) and memory-consistency models.
• Expertise in RTL design (SystemVerilog) of routers, arbiters, buffers, and asynchronous/CDC interfaces.
• Hands-on performance modeling and analysis — cycle-accurate simulation, traffic modeling, bandwidth/latency closure.
• Experience with commercial or in-house NoC generators/IP and their integration methodology.
• Familiarity with high-bandwidth memory (HBM) and chiplet interconnect (UCIe) traffic characteristics.
• Proficiency in scripting (Python/Perl) for performance analysis and verification support.
• Strong micro-architecture, specification, and documentation skills.
• Demonstrated technical leadership and cross-team collaboration.

Preferred Qualifications

• Experience with multi-plane or 3D-stacked NoC architectures.
• Background in AI accelerator or GPU dataflow interconnects.
• Familiarity with formal verification of coherency/deadlock properties.