On-chip telemetry for system-level AI power analysis

ProteanTecs, based in Israel, has introduced an innovative test system that integrates embedded on-chip telemetry with a machine learning-driven analytics engine. This system is designed not only for system monitoring but also for making necessary adjustments. The embedded monitoring system utilizes on-chip AI agents to gather comprehensive data throughout the production lifecycle, starting from development and extending into system operation. These agents have the ability to access parameters that are typically not available to production test systems, such as power integrity, thermal conditions, and assembly faults.

By combining this data with a cloud-based analytics platform and edge-deployed machine learning models, the system enables inline test decisions from new product introduction (NPI) all the way to high-volume manufacturing (HVM). It is particularly useful in applications like AI data centers. This approach aids in identifying root causes of issues, offering crucial feedback to design and production test teams, and ensuring better system readiness for volume production. During mass production, the AI agents provide continuous cloud-based monitoring.

One of the key features of this system is its capability to optimize VDD settings by utilizing AI models to dynamically adjust the voltage setting during production testing. This allows for tailoring the voltage rail to real-world system workloads and environmental conditions, thereby measuring timing margins under functional loads and making necessary voltage adjustments to reduce power consumption.

According to Evelyn Landman, co-founder and CTO of proteanTecs, “Our embedded hardware monitoring system acts as a sophisticated monitor for the system, capturing critical telemetry. Together with a dedicated software stack, system quality, power consumption, and performance are significantly enhanced. We provide unprecedented deep parametric visibility during PCB, Module, and System functional testing – under actual workloads and configurations.”

Uzi Baruch, Chief Strategy Officer at proteanTecs, emphasized the importance of bridging the silicon-system gap through telemetry. He stated, “This telemetry reveals how the chip behaves within the system context, not just in isolation. It offers a new layer of data – not derived, not inferred, but monitored directly from within the chip.” Baruch also highlighted the necessity of transforming traditional system production methods to adapt to the rapid adoption of complex AI and hyperscale computing. The goal is to shift from a static, reactive process to a predictive, data-driven workflow that can adjust to each device’s actual behavior, enabling precision performance tuning and quality assurance.