Below is a msg I sent to the TCAD TWG during the 97 Roadmapping process. Although the Roadmap is intended to be a list of "technical" needs, the more serious limitations to TCAD's progress are related to cooperation, tech transfer, setting proper expectations, etc. I was only partially successful in getting these ideas into the Roadmap. My goal for next time is to remove the word "predictive" from the Roadmap.

After thinking about the current state of TCAD some more, I now think that we have as many "political" challenges as "technical" ones. Here is my new list of TCAD needs, with an emphasis on the "political" or "philosophical" problems. I think these are worth discussing up-front, before we get too entangled with the more detailed needs of each of the modeling groups (e.g., feature, device, circuit, etc.)

Modeling and Simulation

There are two "difficult challenges" where modeling and simulation can have an impact. These are (1) new materials (e.g. Lo-k) and (2) process control. The first requires a paradigm shift in TCAD, and is the most important need. The inherent limit to any simulation is that you can only model things that you already know and understand. An analogy: if we wanted to do weather forecasting on Jupiter, we already have the simulation tools. The real problem would be that we would first need to collect a large amount of data. TCAD is always chasing the latest technology, and is operating in a sub-optimum mode. The new paradigm that is necessary is for closer cooperation between model developers and technology developers. As a new technology is being developed, the modeler should be there, right alongside. There needs to be better recognition of the inherent value in simulation by technology developers. The characterization of a new material can take years, and the better we understand that material, the sooner it can be used in production.

Another area which could easily cause the industry to fall off the productivity curve is in process control. Current simulators can do an adequate job of telling us WHAT manufacturing tolerances we need to achieve in order to obtain the desired electrical tolerances (e.g., Lgate and gate oxide thickness control for a given Vt control). But they offer little advice on HOW to achieve them. Equipment (reactor) level simulations could suggest hardware improvements. There has already been some work in this field, but much more is needed. The equipment model would be developed in conjunction with the tool. This could greatly impact the way tools are developed.

There is a strong need for more rapid introduction of new models. Universities remain the primary source of new models, and there can be a 2-3 year lag before they are introduced into commercial packages. There are several reasons for this. One is the fact that the largest semiconductor manufacturers have internal TCAD development programs, while the rest of the industry relies on software from a handful of relatively small TCAD vendors. The net effect of this is that it is difficult to come to agreement on what the development needs are. Other TWGs do not share this problem. There is also a fear of helping the mainly U.S. based TCAD vendors too much, as improvements in their software will promptly be shipped overseas. However, we do not have similar concerns in other arenas (e.g., ion implantation equipment). A second reason is the lack of a common development platform. Software developed at a university may not readily fit into vendor code. Vendors may not have the resources to evaluate, let alone implement, every newly developed model. There is no consensus as to which models are best, and we end up with several models that cover the same area. There has been much talk over the need for development platforms, but little progress to date. A third reason is that we still do not have good methods for handing-off results from one complexity level to the next. Universities and vendors suffer also from a lack of state-of-the-art data. In short, there is a bottleneck between model development and implementation, with the result of either a significant time delay, or the model never getting implemented at all.

Modeling and simulation has strong metrology needs. Accurate 2D dopant profiling remains a high priority from previous roadmaps, despite the progress that has been made here. Most methods focus on direct measurement, but it is worth mentioning that inverse modeling can also be used (inferring the dopant profiles from electrical and simulation results). Defect metrology is increasingly important. We also need highly accurate linewidth and oxide thickness measurements on actual devices.

Model calibration could be assumed to be an integral part of the model development. However, this has not been the case to date, and more formal methods for calibration and parameter estimation are needed. Unlike SPICE model development, new TCAD models are proposed without a recommended method for determining the model coefficients. The interactions between process steps (e.g. implant and diffusion) can make it difficult to determine the true cause of an observed effect. When trying to calibrate a simulator to measured results, there are too many tuning knobs available. Also, there have not been enough round-robin studies to show how much variation there is from machine to machine.

Lithography simulation is not mentioned as there is little customer pull for this activity.

Recapping, the highest priority modeling and simulation needs are:

#1) Model and technology co-development (paradigm shift).
#2) Equipment level modeling (with equipment vendors).
#3) Rapid commercialization of new models (Univ and National Labs).
#4) Metrology (2D dopant profiling, actual devices).
#5) Calibration methodology.

Michael Duane
2/3/97