PhD Thesis

< back to PhD thesis

« Built-in test in RF circuits using non-intrusive sensors ».

Author: A. Dimakos
Advisor: H. Stratigopoulos
Co-advisor: S. Mir
President of jury: P. Ferrari
thesis reviewer(s): H. Aboushady, P. Descamps,
thesis examinator(s): M. Bucher,
These de Doctorat Université Grenoble Alpes
Speciality: Nanoélectronique et Nanotechnologies
Defense: March 29 2016
ISBN: 978-2-11-129210-9


This thesis addresses the high-volume production test problem for RF and millimeter-wave (mm-wave) circuits. Testing the RF/mm-wave functions of systems-on-chip (SoCs) incurs a very high cost. Built-in test (BIT) is a promising alternative to facilitate testing and reduce costs, but it is challenging since it should by no means degrade the performance of the Circuit Under Test (CUT). In this work, we study a built-in test technique which is based on non-intrusive variation-aware sensors. The non-intrusive property is very appealing for designers since the sensors are totally transparent to the design and, thereby, the test is completely dissociated from the design. The non-intrusive sensors are dummy analog stages and single layout components that are copied from the topology of the CUT and are placed on the die in close physical proximity to the CUT. They simply offer an “image” of process variations and by virtue of this they are capable of tracking variations in the performances of the CUT. In essence, the technique capitalizes on the undesired phenomenon of process variations. The alternate test paradigm is employed to map the outputs of the non-intrusive sensors to the performances of the CUT, in order to replace the standard tests for measuring the performances directly. The proposed test idea is applied to two different CUTs, namely a 2.4GHz CMOS 65nm inductive degenerated Low-Noise Amplifier (LNA) and a wide-band mm-wave 60GHz CMOS 65nm 3-stage LNA. We demonstrate that by adding on-chip a few non-intrusive sensors of practically zero area-overhead and by obtaining on these non-intrusive sensors DC or low-frequency measurements, we are able to track variations in all performances of the CUT with an average prediction error lower than one standard deviation of the performance and a maximum prediction error that is lower or at least comparable to the measurement and repeatability errors in a conventional Automatic Test Equipment (ATE) environment.

pdf pdf