The results of applying a fully-scalable parallel iterative matrix solver to create a three-dimensional simulation of current flow in a bipolar transistor are shown. The transistor was simulated on the Intel Delta at Caltech using software developed at Stanford University. This simulation is the largest yet in this field and represents the state of the art in three-dimensional device modeling. Such modeling efforts enable process design in the "Virtual Factory."
Semiconductor manufacturing plants today are characterized by high capital costs and low flexibility. They are built to serve one or two generations of technology and usually a narrow product base. New technology development is very expensive and largely empirically done. To be competitive in such an environment requires high capital investment.
In the 1970s, chip designers faced similar investment problems because new designs were largely hand crafted. As a result, few VLSI (Very Large Scale Integration) applications were viewed as being able to justify design costs. The CAD (Computer-Aided Design) revolution changed this situation by empowering a vastly increased number of creative people with the ability to design chips at low economic risk. The CAD revolution also provided first pass success in most cases on new designs.
A current view, widely held, is that a similar revolution using the same general approach can revolutionize manufacturing. The concept is to build a highly flexible computer controlled manufacturing facility; a "Programmable Factory." In parallel with this factory, a suite of simulation tools would be constructed; a "Virtual Factory" capable of emulating all functions of the real factory. Both of these facilities would be controlled and integrated through a Manufacturing Automation Framework (MAF). The comparison of actual results from the real factory with predicted results from the Virtual Factory provides a means for improving models.
One example of work in progress is ongoing at Stanford University supported by ARPA. Here, the overall aim is to implement a basic version of these constituents in a working IC (Integrated Circuit) facility and to demonstrate the power of this approach in developing and applying new technologies. Much of the work in the first two years of this project has focused on developing specific software tools needed for the Virtual Factory such as simulators and design tools; developing the software needed for the Automation Framework including information storage methods and frameworks for linking software tools; and putting into place the hardware and software tools needed to connect the Virtual Factory to the Programmable Factory.
Recent accomplishments include an integrated demonstration of the concept including operational prototypes of many of these tools. More than 20 different projects in the research program were integrated together to show the power of connecting the Virtual and Real Factories to an audience of more than 100 people from industry, government and universities. Scalable algorithms, developed on a workstation, were used to demonstrate an extremely large three- dimensional device simulation; a major new software tool, SPEEDIE, which simulates thin film etching and deposition steps in IC manufacturing, has been released to industry; and a new multiprocessing reactor designed to perform multiple manufacturing steps in a single machine is being used to test new manufacturing concepts in the real factory.