An Accurate, Continuous, and Lossless Self-Learning CMOS Current-Sensing Scheme for Inductor-Based DC-DC Converters

Abstract

Sensing current is a fundamental function in power supply circuits, especially as it generally applies to protection and feedback control. Emerging state-of-the-art switching supplies, in fact, are now exploring ways to use this sensed-current information to improve transient response, power efficiency, and compensation performance by appropriately self-adjusting, on the fly, frequency, inductor ripple current, switching configuration (e.g., synchronous to/from asynchronous), and other operating parameters. The discontinuous, non-integrated, and inaccurate nature of existing lossless current-sensing schemes, however, impedes their widespread adoption, and lossy solutions are not acceptable. Lossless, filter-based techniques are continuous, but inaccurate when integrated on-chip because of the inherent mismatches between the filter and the power inductor. The proposed G_M-C filter-based, fully integrated current-sensing CMOS scheme circumvents this accuracy limitation by introducing a self-learning sequence to start-up and power-on-reset. During these seldom-occurring events, the gain and bandwidth of the internal filter are matched to the response of the power inductor and its equivalent series resistance (ESR), effectively measuring their values. A 0.5 mum CMOS realization of the proposed scheme was fabricated and applied to a current-mode buck switching supply, achieving overall DC and AC current-gain errors of 8% and 9%, respectively, at 0.8 A DC load and 0.2 A ripple currents for 3.5 muH-14 muH inductors with ESRs ranging from 48 mOmega to 384 mOmega (other lossless, state-of-the-art solutions achieve 20%-40% error, and only when the nominal specifications of the power MOSFET and/or inductor are known). Since the self-learning sequence is non-recurring, the power losses associated with the foregoing solution are minimal, translating to a 2.6% power efficiency savings when compared to the more traditional but accurate series-sense resistor (e.g., 50 mOmega) technique.