Invited Talks

INVITED TALK on Wednesday, March 29, 2023

Novel approach to master the challenges of
run-to-run control in high-mix low-volume production – PART 2

Ulf Seidel | Infineon Technologies Dresden

To improve the stability of semiconductor manufacturing processes run-to-run control is used in addition to the internal control-loops of the manufacturing tool. A typical run-to-run strategy comprises two steps. First, a suitable state of the controlled process is estimated using process data of previous materials. In a second step the estimated state is used to calculate the recipe tuning parameters for the next materials. High-mix low-volume production poses a real challenge for the state estimation as data is both scarce and erratic. The estimates can be corrupted in many ways, e.g. by missing, delayed or faulty measurements. In order to avoid yield loss and/or scrap it is vital to validate the states permanently. The validation routines are a crucial part of the control algorithm.

At the previous apc|m conference we proposed to exploit the estimation errors calculated by a Kalman filter for state validation. This year’s talk elaborates on this idea in greater detail. It starts with a brief recap of the main points and implications. Inspired by last year’s audience questions the relation to conventional non-treaded run-to-run control with common state models will be discussed next. We will show that the new approach overcomes some of the main practical disadvantages of the later.

High-mix semiconductor manufacturing requires the estimation of hundreds or even thousands of interconnected states over long time periods (years). This calls for an estimation procedure that is numerically highly stable. Such an algorithm will be developed in the talk step by step. It is based on the generalized Kalman filter proposed last year that copes with the intrinsic ambiguity of run-to-run models and allows automatic state initialization. We will present a solution that shows superior numerical stability while the incurred computational costs are comparable to the conventional Kalman filter.

You are interested in the 1st part of the talk? You can watch it here:


Curricula Vitae Ulf Seidel

Ulf Seidel is Principal Engineer for Advanced Process Control working both in operations and research at Infineon Technologies Dresden, Germany. He received the M.Sc. degree in Control Engineering from the University of Sheffield, U.K., in 1994, and the Dipl.-Ing. degree in Electrical Engineering as well as the Dr.-Ing. degree (Ph.D.) in Control Engineering from TU Dresden, Germany, in 1995 and 2000, respectively. He was with Klippel GmbH, Dresden from 2000 to 2008 where he developed instruments for diagnosis and quality control of audio systems including novel online identification of nonlinear loudspeaker models. He joined the run-to-run control group at Qimonda Dresden from 2008 to 2009. From 2009 to 2010 he was with InfraTec GmbH, Dresden developing measurement systems for pyroelectric infrared detectors. Since 2010 he has been with the advanced process control group at Infineon Technologies Dresden.