Sep 13 (Wed) @ 1:00pm: ”Machine Learning Approaches for Simultaneous mm-Wave Design Automation and Optimization,” Andrea Arias-Purdue, ECE PhD Defense

Date and Time
Engineering Science Bldg (ESB), Room 1001

Zoom Meeting: Meeting ID: 857 4664 8423 | Passcode: 693376


Machine learning (ML) has been successfully demonstrated for power amplifier (PA) matching network (MN) design, non-linearity and bias compensation as well as for linear and nonlinear device modeling.  However, a device to PA framework that simultaneously automates and optimizes the PA for a set of target specifications such as power, efficiency and linearity is not currently available. Above 100 GHz, transistor models are typically poorly validated, making PA design a fairly trial and error field, often requiring 1-2 design spins before achieving target performance. This work focuses on ML and automation techniques for device modeling and MN design.  Part I introduces a traditional PA design flow and reports state of the art output power to die area and power-added efficiency (PAE) at mm-Wave.  In Part II, power dataset-aware Artificial Neural Network (ANN) models that predict transistor performance at D-band (110-170 GHz) are presented, demonstrating 2X-5X improvement in measurement agreement as compared to existing device models.  The ANN modeling methods are applied to 250-nm InP HBTs and 40-nm GaN HEMTs.  Two GaN HEMT embedding networks are systematically analyzed for gain-boosting in Part III, resulting in state-of-the-art gain and power utilization factor device-level performance.  MN designs that maximize reflection coefficient margin, PAE and output power across a 10% bandwidth centered at 140 GHz are presented in part IV.  This work provides insights into multi-variate comprehensive analyses and ANNs for PA design, presenting progress towards mm-Wave design automation and rapid optimization.


Andrea Arias-Purdue holds a B.S. and M.S. in physics from California State University, Fresno and a M.S.E.E from the University of California at Santa Barbara (UCSB). She is currently a Ph.D. candidate at UCSB in Prof. Buckwalter’s RF & Mixed-signal Integrated Systems Laboratory Group. She has over 10 years of experience in high-power semiconductor design and yield improvement for high frequency applications, including novel N-polar GaN devices, as well as III-V power amplifier MMIC design.

Hosted by: Professor Jim Buckwalter

Submitted by: Andrea Arias-Purdue <>