A Nonintrusive Machine Learning-Based Test Methodology for Millimeter-Wave Integrated Circuits

Abstract

In this manuscript, we leverage the power of machine learning algorithms to propose a test methodology for mmwave integrated circuits. The proposed test strategy is based on identifying the main process degradation mechanisms in a particular Device Under Test (DUT) and then designing dedicated process monitor circuits to characterize this degradation and infer the DUT performance. The resulting process monitors do not load or couple to any of the DUT nodes and the methodology can be adapted to any mm-wave device without complex co-design. The proposed test methodology is illustrated on a set of 21 fabricated samples of a 65 GHz PA designed in STMicroelectronics 55 nm CMOS technology.

Built-In Self-Test of Millimeter-Wave Integrated Front-End Circuits: How Far Have We Come?

“…A simulation study shows that the approach is able to extract the linear and nonlinear performance parameters of the LNA with sufficient accuracy. In [39], a similar approach is used to test a 65 GHz-PA in 55 nm-CMOS. In contrast to [38], the approach is verified with a fabricated batch of 21 samples.…”

Nonintrusive Machine Learning-Based Yield Recovery and Performance Recentering for mm-Wave Power Amplifiers: A Two-Stage Class-A Power Amplifier Case Study

“…In this paper, we extend our previous work by demonstrating the feasibility and performance of this technique with the design and experimental characterization of an integrated proofof-concept prototype featuring the proposed calibration technique. Compared to the state-of-the-art, the proposed prototype combines the optimized nonintrusive performance monitoring developed by the authors in [22] with the use of low design complexity and generic tuning knobs that do not require modification of the circuit power supply. In this regard, this paper presents a 69 GHz PA with embedded one-shot calibration and non-intrusive performance monitoring.…”

“…In this paper, to guide the calibration algorithm, we adapt the nonintrusive test methodology proposed by the authors in [22]. In this regard, we propose to integrate a set of nonintrusive process monitors together with the circuit under calibration.…”

“…These process monitors are stand-alone circuits, not connected to the circuit under calibration, that are designed to output DC or low-frequency signatures sensitive to the variations of a selected set of physical parameters in the fabricated silicon. By targeting the physical parameters critical for the performance of the circuit, it is possible, as demonstrated in [22], to train a machine learning model to predict the performance of the fabricated circuit from the process signatures provided by the monitors. In this work, and given that the process signatures are strongly correlated to the circuit performances, we propose to employ this nonintrusive on-chip monitoring strategy to guide the calibration algorithm and provide the signatures to train the regression function in (1).…”

“…These process monitor circuits have to be specially designed to be sensitive to the main performance degradation mechanisms of the PA. Authors presented in [22] a systematic methodology for the design of these process monitors in the context of mm-wave circuit testing. The design methodology for the proposed calibration application is a natural extension of the one presented in [22].…”