System Architect
TLDR
Own the SoC and product specification stack for a 1.6T retimer SoC and 224G/lane PAM4 AEC product, defining normative specs across RTL, firmware, software, and qualification.
At Kandou, we are redefining the economics of AI infrastructure. Our mission is to democratise AI by significantly reducing the Total Cost of Ownership (TCO) of hardware systems — a critical barrier to scalable adoption.
Our proprietary MIMO-over-copper technology powers a high-performance, chiplet-based AI memory fabric that is both scalable and energy-efficient. Unlike traditional interconnects, our solution reduces power consumption significantly while preserving high bandwidth and ultra-low latency — unlocking unprecedented efficiency for AI training and inference at scale.
Kandou’s architecture is not just an incremental improvement — it’s a foundational shift in how AI hardware is built for the future.
We are actively seeking a System Architect based in US (Bay Area or Austin preferred), EU considered.
- Own the SoC and Product Specification stack for a 1.6T retimer SoC and the AEC product it enables, at 224G/lane PAM4.
- You will author and maintain normative specifications (SHALL/MUST/SHOULD/MAY language with REQ-ID traceability) that RTL, firmware, software, SI/PI, manufacturing, qualification, and customer engineering teams execute against.
- Standards scope spans IEEE 802.3ck/dj, OIF CEI-112G/224G, OSFP MSA, CMIS 5.x, EIA-364, JEDEC JESD22/51.
- You will represent architecture in hyperscaler and strategic platform customer specification negotiation, including qualification test point, FEC telemetry, CMIS vendor-page, and thermal envelope commitments.
Required: - 12+ years in shipping high-speed interconnect (retimer, gearbox, PHY, AEC, AOC, coherent optics, SerDes-centric networking silicon).
- Prior System Architect, Chief Architect, or Principal Systems Engineer role on a shipping 100G/lane or higher product.
- Direct authorship of at least one product spec and one silicon spec on a shipping product.
- PAM4 link training, FEC, and on-die telemetry fluency at the specification level.
- DOORS / JAMA / equivalent traceability tooling.
KandouAI is transforming AI infrastructure by reducing the Total Cost of Ownership of hardware systems, making scalable AI adoption more accessible. Our proprietary MIMO-over-copper technology delivers a high-performance, chiplet-based memory fabric that is both energy-efficient and ultra-fast, fundamentally changing the way AI hardware is constructed for the future.