Combining Internal Probing with Artificial Neural Networks for Optimal RFIC Testing

Abstract

In order to reduce production costs of RF devices, it is important to remove bad circuits very early in the production flow. It is all the more true for dies designed to be integrated in complex systems. Thus highly efficient RF wafer testing is mandatory for those applications to prevent the loss of assembled systems due to defective RF dies. The problem is that current RF probing technologies hardly fulfill the industrial test requirements in terms of accuracy, reliability and cost. The proposed method proves to be a very interesting alternative to validate RF parameters with no need of expensive RF equipments (RF probes and RF automated test equipments (ATE)). A new test strategy based on DC or very low frequency (LF) measurements, which allows the elimination of expensive RF tests, is presented. The main idea is to insert some simple design for test (DfT) circuitry within the chip.This DfT provides relevant information on the structural behavior of the device blocks. The internal node data are additional to standard DC test measurements like power supply current or advanced DC test signatures (e.g. Vdd ramping), and LF measurements like gain in loopback mode.Since RF performance of each block is directly related to such structural data, it is possible to predict the RF characteristics of the blocks without time consuming RF measurements. RF parameters estimation is performed using nonlinear Artificial Neural Networks.

Parametric Built-In Test for 65nm RF LNA Using Non-Intrusive Variation-Aware Sensors

“…From (1), we can write To this end, we can use the alternate test paradigm to infer implicitly the performances from the sensor measurements [13,16,25,30]. Specifically, in an off-line preparatory training phase where we employ a large set of circuit instances with process variations, we learn one regression function per performance that maps the sensor measurements to the performance:…”

“…Another popular builtin test technique relies on the use of envelope detectors [6,7,18,29,32] and current sensors [11,15] to extract DC/low-frequency test signatures that nevertheless carry RF information. Finally, alternate test is a generic technique that aims to map low-cost measurements to the RF performances [13,16,25,30], thus circumventing the need to measure directly the RF performances. Alternate test can also rely on information-rich, low-cost measurements obtained by built-in sensors [1].…”

Implementing model redundancy in predictive alternate test to improve test confidence

“…Alternate testing has been widely studied in the literature for many years [1,2]. Despite the number of good results reported in all these works that illustrate the efficiency of the alternate test strategy, its industrial deployment is still limited, mainly because of a confidence problem.…”

Estimation of RF PA Nonlinearities after Cross-correlating Current and Output Voltage

“…The lack of harmonized and universal RF structural tests have forced industries to resort to specifications' functional testing in order to ensure the required low defect ppm (ideally zero) quality level, eventually relying on both builtin and built-out tests [4,14].…”