On-Die Droop Detector for Analog Sensing of Power Supply Noise

Understanding the supply fluctuations of various frequency harmonics is essential to maximizing microprocessor performance. Conventional methods used for analog validation of the power delivery system fall short in one or more of the following areas. 1) Measurement accuracy in both frequency and time domains, especially for very high-frequency noise caused by large di/dt events. The multigigahertz power supply noise attenuates very quickly away from the die. Conventional approaches of measuring the noise at the pins of the package or at the die using capacitive probes are not accurate for multigigahertz clocks. For this reason, the observability of high-frequency on-die noise has been very tricky. 2) Implementation (e.g., delivery) of analog references to multiple areas across a "noisy" die, compactness/modularity of the measurement units, restraining assumptions inherent in the measurement circuit such as periodicity of the supply noise event. 3) Automation to enable a timely volume of measurements. The efficiency of the measurements is key to correlating a particular speed path to poser supply noise. To address these issues, this paper presents an on-die droop detector (ODDD), a scalable IC solution implemented and validated on a 90-nm process, for analog sensing of differential high-bandwidth supply noise.