weeteq

Mr. Martin MacDonald, Co-founder & Chief Operating Officer, weeteq

Martin MacDonald is Co-founder and Chief Operating Officer of weeteq, a fabless semiconductor company building the power-delivery layer for gigawatt-era AI data centers. weeteq's Ultra Edge® predictive transient-correction silicon and DIVA™ 800V end-to-end architecture together deliver approximately 30% more tokens per watt — work recognized with Best Paper in the AI/HPC category at the OCP EMEA Future Technologies Symposium in Barcelona.

Martin leads weeteq's commercial and operational functions, spanning fundraising, partnerships, supply chain and the company's engagements across the AI power chain, including tier-one xPU vendors, server ODMs and power semiconductor manufacturers. 

He co-founded the company with Dr. Taner Dosluoglu and built its mixed-signal design team in Glasgow alongside a second design center in Istanbul.

Before weeteq, Martin spent over two decades in enterprise technology, with roles at Honeywell, Dialog Semiconductor, Cisco, Sabio and STAC. His background spans commercial leadership, operations and go-to-market for silicon and systems businesses across Europe, North America and Asia. He is based in Edinburgh, Scotland.

 

Topic:

Increasing tokens per Watt by 30%, without compromising efficiency

 

Abstract:

Power delivery solutions within AI and high performance compute (HPC) data centers have received significant recent attention, with a higher voltage input bus being planned for server trays. In parallel with this effort, high performance processor manufacturers integrate on-chip solutions to address the power delivery network challenges of fast load transients within power domains. Higher power requirements have necessitated higher voltage components to be placed closer to the package, while the efficiency requirements have limited the available solutions with high bandwidth to provide active droop compensation. 

weeteq offers a different approach. Ultra Edge® is a microcontroller-class predictive controller that corrects load transients in the µs-to-ns range, ahead of the step rather than after it. DIVA™ is a parallel dual-path architecture carrying 800V DC end to end to the point of load, eliminating conversion stages rather than optimizing within them. Together they deliver approximately 30% more tokens per watt — work awarded Best Paper in the AI/HPC category at the OCP EMEA Future Technologies Symposium.

This ten-minute session covers the physics of the transient problem, the dual-path architecture and its control layer, first results, and what a 1,000A demonstrator means for 800V rack-scale deployment. Because weeteq operates at the control and architecture layer above GaN and SiC devices, it makes device vendors and ODMs partners rather than competitors.

 

Key Technologies Covered:

  • Predictive load-transient correction (µs-to-ns response)
  • Ultra Edge® microcontroller-class controller IC
  • DIVA™ parallel dual-path power delivery architecture
  • 800V DC end-to-end distribution to point of load
  • Fast DCM high-density conversion stage
  • Rack-scale HVDC power distribution for 1MW+ AI racks
  • Guard-band and over-provisioning reduction in accelerator power domains
  • High-slew dynamic voltage scaling for xPU workloads
  • Control-layer integration with GaN and SiC power devices
  • Mixed-signal implementation on TSMC 22nm