On-chip telemetry for system-level AI power analysis
ProteanTecs in Israel has launched a test system to combine embedded on-chip telemetry with an ML-driven analytics engine that can also be used for system monitoring and adjustment.
The embedded monitoring system uses on-chip AI agents to capture deep data throughout the production lifecycle, from development and into system operation. The agents access parameters used in functional test that are not accessible to production test systems such as power integrity, thermal, and assembly faults.
This data is combined with a cloud-based analytics platform and edge-deployed ML models for inline test decisions from new product introduction (NPI) to high-volume manufacturing (HVM) and in use in applications such as AI data centres. This helps to identify root causes, providing critical feedback to design and production test teams and ensuring better system readiness for volume production. During mass production, the agents provide ongoing, cloud-based monitoring.
- Anticounterfeit chip monitoring at BAE Systems
- GUC, proteanTecs team for die to die interconnect
- When AI agents fight
The system enables VDD setting optimisation by using AI models to dynamically refine the voltage setting during production testing, tailoring the voltage rail to real world system workloads and the system environment. This measures the timing margins under functional loads and adjusts the voltage settings, reducing the power consumption.
“Our embedded HW monitoring system serves as a sophisticated monitor for the system, capturing critical telemetry. Together with a dedicated software stack, system quality, power consumption and performance are significantly improved,” said Evelyn Landman, proteanTecs co-founder and CTO. “We deliver the first-ever deep parametric visibility during PCB, Module, and System functional testing – under actual workloads and configurations.”
“Bridging the silicon-system gap, this telemetry reveals how the chip behaves within the system context, not just in isolation,” he said. “It provides a new layer of data – not derived, not inferred, but monitored directly from within the chip.”
“Today’s rapid adoption of complex AI, hyperscale computing is stretching traditional system production methods beyond their limits,” said Uzi Baruch, Chief Strategy Officer at proteanTecs. “We’re transforming test from a static, reactive process to a predictive, data-driven workflow that adapts to each device’s actual behaviour. This empowers teams to move beyond oversimplified worst-case assumptions and into an era of precision performance tuning and quality assurance.”
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
