RESEARCH / 2026.09
Research programme
Six long-term programmes connect semiconductor manufacturing, lifecycle carbon, physical telemetry, AI workloads and place constraints into one public measurement architecture.
R01
Workload energy attribution
Physical telemetry → workload evidence.
409.7 Wreference signal
↗
R02
Lifecycle carbon of compute
Silicon → hardware → facility → operation → useful compute.
LCfull carbon boundary
↗
R03
AI inference intensity
Task, batching, hardware and precision.
25×energy / response
↗
R04
Water as a binding constraint
Physical water use × local basin context.
>80%extreme stress
↗
R05
Semiconductor manufacturing carbon
Fab → wafer → good die → package → useful compute.
CL-SCMmanufacturing method
↗
R06
Time & region shifting
Constraint-aware workload placement.
50%+location potential
↗
02 / Programme system
Each programme is a continuing research line, not a single article.
Research questions lead to active projects; projects release methods, datasets, working papers and assessments.
Programme
Research question
Active project
Expected public output
R01How should physical energy be attributed to delivered compute?CL-P01 / Compute Energy Measurement ProtocolCL-WP-005 v0.9 · UAWM method · primary campaign next
R02What is the full lifecycle carbon intensity of useful compute?CL-P02 / Carbon of ComputeWorking paper · lifecycle schema · benchmark set
R03How do task, batching, hardware and precision change AI energy?CL-P01 / inference benchmark streamCL-WP-005 v0.9 · task-normalized energy accounting
R04When does local water availability become the binding compute constraint?CL-P03 / Place Water Constraint ModelModel note · place evidence requirements
R05How much carbon exists before an accelerator becomes operational?CL-P02 / semiconductor workstreamCL-SCM method · public benchmark bridge
R06Where and when should flexible compute run after physical constraints?CL-P05 / Constraint-aware RoutingResearch note · decision records