Interview with Mark Law

Mark Law is finally a professor at the University of Florida. He needs no further introduction:

You recently moderated a panel discussion on how to attract students to semiconductors. Can you comment on that subject?

Semiconductors are increasingly seen as a non-sexy research area by incoming grad students. They want to work on internet, communications, biotech... The semiconductor industry is seen as corporate America - no real exciting opportunities. People are making millions (billions?) with software companies in garages. Kind of like people used to start semiconductor and computer companies. After all, we have a roadmap that lays out everything for the future.

The roadmap is a negative for recruiting students - they can see themselves as a cog in the wheel. We need to do a better job of communicating the fun stuff that is "off roadmap" or "at the end of the roadmap". That's where excitement should lie for a new student. What happens to sub 10nm transistors - how do we make them and what do they look like?

The TCAD Journal has fast turn-around time, no page limits, color, and unlimited distribution (WWW). Yet many authors still prefer to submit to more established journals first. What can we do about this?

You left out a friendly editor :-)

The biggest problem is readership. We have a hard time quantifying it. We can count hits, but that tends to be pretty small. Journals get subscriber numbers. I don't believe that the number of subscribers who read a particular article are much different than hit count we get on a web page, but the potential is there.

I think we need to start a subscriber service. We can email announcements to a list when papers are accepted. The email would contain the URL and an abstract say. We could then point to the number of names on the email list as our "subscriber" base.

Do you have a favorite TCAD story?

PISCES originally stood for PoIsson and Single Carrier Equation Solver. Mark Pinto made a bet he could add a second carrier to PISCES in a weekend, and the race was on. He didn't win the bet, but if I recall correctly, he came awful close. For those of you who have worked on PISCES, that explains a lot - doesn't it?

[A trivia question posed by Mark:] So why is the two.dim model called that in SUPREM-IV?

What are some of the things you see on the horizon for TCAD?

The main developer of Pisces II (Mark Pinto) discarded it and rewrote it. He uses PADRE to do device simulation. The main developers of SUPREM-IV (Conor Rafferty and Mark Law) discarded it and rewrote it. They use PROPHET and FLOOPS, respectively, for process simulation. Why are codes with direct lineage to PISCES and SUPREM-IV still pretty much industry standards, when the original developers decided a rewrite was necessary?

What happens next? I hate to get the crystal ball out, but... Let me limit myself to process simulation, since that's where I am best qualified to make wild guesses.

I would say that traditional "model" development will get deemphasized. Characterization and fundamental approaches will dominate over traditional TCAD. We'll be doing more ab-initio, Monte Carlo, and molecular dynamics in research. This will be combined with new experimental work and techniques to help verify and parameterize models. The research community will nearly "retire" from the world of continuum simulation - we'll just use it to verify ideas developed in other realms.

and, can you explain the case of the tunneling implant?

Sure... Suprem IV uses dose matching to get implant offsets for multiple materials. This means the code has to compute how much dose was in the top layer. Then an equivalent thickness of the second layer is computed that would have stopped this much dose. This is then used as a depth offset in computing the second layer implant.

For example, if I have 100A of oxide on silicon, it would compute the dose in the oxide (say 1e12). It then computes the amount of silicon required to stop 1e12 dose, say 80A. All the depths in the silicon are made 80A deeper.

I used a fixed integration method in Suprem IV to compute the dose (you have to integrate the profile, after all). The spacing of the integration was 5A, if I recall correctly. If you integrate the dose in the top poly with this operation in your example, you probably get something close to 99.99% of the dose. The beauty (or inherent evil) of log scales is that isn't good enough! You then get a tail (because 100% of the dose isn't stopped in the top material). If you increase the energy, the integration gets more accurate (the profile is spread broader, and a fixed integration gives greater accuracy) and away goes the tail.

I fixed it in FLOOPS by using an adaptive integration (it changes grid spacings to get desired accuracy).

do you want to say anything about equation interfaces yet?

TMA's process module equation interface (PMEI) has gotten a lot of buzz lately. It's what the industry has claimed to have wanted for years. It was definitely time for a rewrite (see above), but I'll be very interested to see who actually uses the code to develop their own models.

It's a lot harder to write a model that works than people think. There are always convergence issues. There are always unexpected interactions with the other models. There are going to be big support issues with these things.

When we coded a {311} defect model for the first time, it was hard to keep it from undershooting. The dissolution rate can be very high, and the time integration method can easily allow the concentration to go negative. Bear in mind, the equation I'm talking about is just:

dC/dt = -kC

This is not the most difficult equation to solve, but normal PDE solvers don't deal with rapidly decaying values very well. Most of the equations we write don't have a steady state solution of zero. This can be a very stiff problem in time, and careful control has to be exercised over the time step. My guess is that the PMEI code doesn't expose the time step selection algorithm.

Having said all that, it will make my life easier. I can do tech transfer to the commercial codes a lot more easily. The papers we write can be used to implement the model directly with little work. This should mean the modeling we do will appear quicker in commercial codes.

Mark Law, Professor                     Phone: (352) 392-6459
Director of Undergraduate Programs      Fax: (352) 392-8381
Department of ECE                       email: 
University of Florida
535 New Engineering Building
Gainesville, FL 32611-6130

Home * Calendar * Calibration * Publications and Presentations * Editorial/Humor * Miscellaneous

This page last updated Jan 16, 1999 by