Meta's Infrastructure Silicon organization designs custom silicon that powers our data center infrastructure - SmartNICs/IPUs/DPUs, AI accelerators, and networking ASICs. We are looking for a Data Processing Unit (DPU) I/O Architect to own the I/O subsystem and the hardware/software interface between our silicon and the software that drives it - PCIe, DMA and offload engines, I/O virtualization, Confidential Computing, and system management. You'll define how the hardware presents itself to firmware and software - the register and programming models, descriptor and doorbell interfaces, and offload interfaces - and decide what belongs in hardware versus firmware versus software. You'll work hand in hand with the firmware, driver, and RTL teams, and carry your designs from early path-finding through implementation and silicon bring-up.
Responsibilities
Own the I/O subsystem architecture for our DPUs - PCIe, DMA and offload engines, I/O virtualization, confidential computing, and system management -- from early path-finding through silicon bring-up
• Decide what runs in hardware, what runs in firmware, and what stays in host software, and design the interfaces between them
• Define the hardware/software interface - register and programming models, descriptor and doorbell interfaces, error and interrupt models, and the offload interfaces firmware and drivers build on
• Work closely with firmware, driver, and software teams so the interface is clean, efficient, and genuinely usable, and evolve it as their needs change
• Weigh the performance, power, and area trade-offs behind your choices and back them with data, partnering with performance modeling where deeper analysis is needed
• Work closely with design, DV, and PD teams on feature definition and PPA tradeoff; refine architecture to meet the design and PD constraints. Define architecture to maximize DV complexity including defining specific features to help ease DV
• Support Design, DV and PD to resolve interface and integration issues as they come up. Support post-silicon bring-up of the I/O subsystem
Minimum Qualifications
• Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience
• 8+ years of relevant industry experience architecting I/O and system subsystems and their hardware/software interfaces - register and programming models, DMA and descriptor interfaces, doorbells, and offload interfaces - for NIC/DPU or comparable ASICs
• Experience with PCIe data link and transport protocols, including TLP formats, transaction types (posted/non-posted/completions), and ordering rules
• Experience with I/O virtualization technologies such as SR-IOV and SIOV
• Experience with PCIe capabilities such as Advanced Error Reporting, Downstream Port Containment, and Integrity and Data Encryption
• Experience with on-chip scheduling and arbitration architectures and quality-of-service mechanisms - transaction/packet schedulers, arbiters, traffic shapers, and rate limiters (meters/policers, token buckets) - and their impact on latency, fairness, bandwidth partitioning, and throughput
• Experience with symmetric and asymmetric cryptography concepts and inline data encryption (e.g., AES-XTS, AES-GCM)
• Experience with x86 and/or ARM system architecture, including virtualization technologies, IOMMUs, and interrupt virtualization
• Experience with data center workloads and their I/O characteristics
• Experience driving architectural decisions independently and influencing their direction through data
Preferred Qualifications
• PhD in Computer Science, Computer Engineering or Electrical Engineering
• 15+ years of relevant industry experience architecting I/O and system subsystems and hardware/software interfaces for NIC/DPU or comparable ASICs
• Experience designing PCIe endpoints and root complexes
• Experience with hardware description languages (e.g., SystemVerilog, VHDL) and simulation environments used in ASIC development flows
• Experience developing Python-based automation pipelines for simulation orchestration, regression testing, and performance data analysis
• Experience building or scaling performance modeling infrastructure for hyperscale data center ASICs, including network, storage, or AI inference accelerator designs
• Experience with system-management concepts including server/BMC interactions, and protocols such as I2C, I3C, USB/Ethernet, USB/MCTP, and SPI/eSPI
• Experience designing high-performance DMA engines and offload accelerators, such as for cryptography and compression
• Experience with post-silicon performance validation and model-to-hardware correlation methodologies
• Understanding of Confidential Computing use cases and the related PCIe protocols -- Data Object Exchange (DOE), Security Protocol and Data Model (SPDM), Management Component Transport Protocol (MCTP), and TEE Device Interface Security Protocol (TDISP)
