Hank Walker
******************************* Duncan M. Walker teaches at Texas A&M University at College Station, Texas, where he's a faculty member in the Department of Computer Science. You've got to know three things before you talk to Prof. Walker. First, he goes by "Hank" and second he's not a Texan by birth. His grandfather had a farm in the San Fernando Valley, with their house on Sepulveda Blvd. in Van Nuys, and hence Walker is a Californian through and through. Be that as it may the third thing you should know before talking to Prof. Walker is that the folks at Texas A&M are Aggies, not Longhorns. The folks at the University of Texas at Austin are Longhorns. If you call Prof. Walker "Duncan" rather than "Hank," he'll probably look past it. If you mistake him for a Texan, he'll probably let than one slide as well. But if you get Aggies confused with Longhorns, Hank Walker may just get up on his saddle and ride off into the sunset. Luckily Walker was in a forgiving mood the day we spoke by phone and we were able to get past those embarrassing Aggie-Longhorn opening moments. After I swore on Sam Houston's grave never to make that mistake again, he agreed to finish the interview. ******************************* Hank Walker came to Texas A&M in 1993. Prior to landing in the Lone Star State, he spent 14 years at CMU Carnegie Mellon University first as a grad student and then as a research faculty member and assistant director of the CAD Center there. Walker also did hard time in industry as working on digital signal processing for night vision at Hughes Aircraft, as a microprocessor designer at DEC, and as an EDA tools researcher at IBM. Walker says, "My primary research focus is IC test and diagnosis, but I'm also working in static timing analysis, and once in a while return to my background in yield modeling. I'm very happy to be at Texas A&M and to have the chance to work here with the great faculty in EDA and test Jiang Hu, Peng Li, Weiping Shi, Sunil Khatri, Gwan Choi, Ray Mercer, and Rabi Mahapatra. In addition, here at Texas A&M we have one of the largest mixed signal research groups in the U.S., along with the strong Analog and Mixed Signal Center. All the way around, Texas A&M's a great school." Walker says his biggest transition in coming to Texas in 1993 was that given his years associated with private universities, Caltech as an undergrad and CMU as a grad student and professor Texas A&M is a public university and that can be challenging for people who teach VLSI design. (More on that later) Public school or private, Hank's been involved non-stop in VLSI design and CAD tool research since he was an undergraduate starting in 1977. He says, "Back then, we were struggling to make graphical layouts with the new tools, and Calma and Applicon were Kings of the Road. I remember Ron Roher's talk from the 25th DAC where he reminded people that back in those days, you were just pushing polygons around, running SPICE or perhaps its predecessors." "To me back then, just having a color monitor was a big innovation because it was so much easier to work in color rather than in black and white. We had a brand new Alto computer where we used Icarus developed by Doug Fairbairn. Also at that time, Carver Mead was developing his book on VLSI design along with Lynn Conway, and we were actually testing out a prototype of the book in my VLSI class at Caltech." I asked Walker if he was ever inclined to write a history of VLSI design automation. He said, "No, because Ron Roher is in a much better position to write that history than I am, given that he's been around for all of it. His was a generation that was young enough that getting in when they did, they saw it all." ******************************* Walker says that undergraduate VLSI design courses have changed a lot since his college days at Caltech "When I was a student, VLSI design was mostly layout oriented. Now, we keep going up abstraction levels to more and more of a system level. In fact, considering the synthesis tools and the cell libraries available today, the question is becoming when is it really VLSI design and when is it not? If you've got an FPGA with IP blocks is that VLSI design? Or is it logic design? Usually, we talk about ASIC design or customizable design, but it used to be that when you said VLSI design, it was about doing a custom or ASIC chip. Today when you're doing design with structured ASICs or gate arrays, the definition of design gets real fuzzy." Meanwhile, Walker says that his students at A&M are using more and more programmable devices in their classes because of the NRE costs and turnaround time associated with ASIC. Although he adds, "In analog design, of course, it's still very much in the same mode that it was in when I was a student. You do hand layout and circuit simulation. A lot of the algorithms used in analog design are in the tools from MatLab, but even then your have to fabricate and test the chips to see if they really work." "There's one other point about fabricated student projects the 8 to 12-week cycle time means it's only an option in a year-long course, or a sequence of linked courses, but not in a one-term or one-semester course." Walker is actually questioning whether or not it's a useful exercise for students to see the chips they've designed in a class actually fabricated and tested. He says that when chips are manufactured from student designs, Texas A&M usually uses MOSIS, although predominantly for the fabrication of analog parts. Alternatively, he says the university will sometimes fabricate designs through a particular company that the school might be working with. For instance, the analog research group at A&M has a close relationship with Texas Instruments. TI might, therefore, arrange to manufacture a student's design at their facilities when appropriate. Prof. Walker says that seeing their designs translated into silicon is not as useful an experience for students as it has been in the past: "If you design some little data path, it will most likely work. Bigger chips, however, are really a problem. The benefits of fabricating and testing these chips particularly as they rarely actually work is just not there. It's not worth the effort. Our labs on campus do fabricate designs, but they're used mostly for fabricating special devices sensors and MEMS, for example." ******************************* Today the courses taught at A&M cover the gamut from design to the development of the tools used for development "Making tools or using tools we teach both subjects. However, designing EDA tools is more of a graduate-level topic because most undergraduates haven't had enough course work in algorithms to grasp the material related to tools development. We do encourage our seniors to focus on system design, which allows them to look at both the process of design and the tools needed to work at that abstraction level." "Texas A&M teaches a general undergraduate EDA course that overviews the fundamentals it's about IC design, test, algorithms, DFT, and so forth plus we have some specialized courses that focus on interconnects, routing, and circuit extraction. These more specialized courses are developed based on the specific interests of the faculty. We feel we give our undergraduates a solid exposure to the tools for design, and our graduate students have extensive exposure." Walker says the most important teaching objective in his department is to instill a sense of reality into students' course work: "From a practical point of view, our students want to get a job after graduation and so we try to help them be up to speed on all the latest tools and techniques. Our challenge really is to teach everything to every student. Some students end up focusing on one thing more than another especially in the senior courses but throughout the entire curriculum we try to present a realistic team environment for design. Many of our students are looking at the EDA tools from a computer engineering perspective, so were also teaching a lot of hardware/software embedded tools to add an additional reality component to those considerations." "How do we keep our students current when the technology is changing so rapidly? Basically, we focus on things that we feel will still be true in 20 years. The fundamental algorithms will still be true, and stochastic optimization will still be a good idea in 20 years. So we continue to teach tried and true ideas in design and EDA, rather than just the latest idea, which might or might not be good by next year even. But we also don't want to just teach about the things that are mature in the industry. We want to talk to the students about the latest things going on in the industry. It's a tricky balancing act." "The stuff they're learning at Texas A&M will be useful on the job for our students software engineering, computer engineering, and so forth. In addition, our students experience some of the practical limitations in designing chips things that in a perfect world they wouldn't have to deal with. For instance, they practice making IP blocks above and beyond using the commercially available IP blocks provided to them but the blocks our students develop are not really sophisticated enough to hand off to another design team, at least not in a commercial sense." "They do their IP designs in Verilog by the way, VHDL is dead and they hand their Verilog code off to another team member. It's interesting to have one set of students creating the IP and having another set of students using that IP. They all quickly understand that if the IP you're using doesn't have a good yield, your design and your company will die. They quickly understand that it's in the interests of the IDMs, the foundries, and all users, that IP be good. Even more quickly, our students learn the benefits of providing a nice, structured model for the IP blocks and the benefits of writing carefully documented code." "We use a lot of programmable devices and IP from various vendors in teaching our courses at A&M. The reality for our students is that when you've got a chip with a billion transistors, you're not going to be fiddling around designing individual transistors. It's important for our students to have that perspective if you have to deal with a huge amount of complexity, how are you going to manage that? So teaching to specific, current EDA tools that are not yet dealing with that complexity is not really my focus it's the long-term play in the industry we want our students to understand via the classroom." ******************************* Nonetheless like professors everywhere, Hank Walker needs to have access to the current set of commercial EDA tools. But the costs and licenses remain problematic. "When I was a graduate student at CMU, I tried to get a copy of Dracula from ECAD for my research. The price for a license was $100,00 plus $12,000 in annual maintenance. They were willing to knock it down to 'just' the $12,000 annual maintenance on each license, but that was back when a graduate student got paid $500 a month, so even that cost was hard to justify." Walker says that situation had an impact on a former office mate of his, John Osterhouse (the developer of Tcl/Tk): "After graduation, John joined the U.C. Berkeley faculty and developed tools widely used in universities, including the Caesar, Magic and Crystal systems. They found wide university use due to the difficulty in obtaining commercial tools, including commercial tools that ran on generic UNIX workstations, rather than vendor-specific platforms. Remember when EDA vendors made most of their money selling hardware?" Today things are different, according to Walker: "We use something from everybody and the prices are adjusted for the universities. We use Design Compiler from Synopsys, and for simulation we use VCS. We use Cadence tools for layout types of things, and routing. Then we use Cadence Spectre for simulation. For things like test generation, we use Mentor's FastScan. Other faculty members here at A&M use TetraMax, and in the past I've used Cadence's Encounter Test. For DRC, we're using Mentor's Calibre. Since we're using Calibre for research and teaching, the friendliness of that particular group at Mentor actually factors in our choosing to go with the Mentor tools. Also, Mentor is cheaper from the standpoint of the university." "Overall however, we need to have a whole bunch of tools from all of the main vendors generally these days, you get all of the tools from a particular vendor for just one license fee. Of course, every vendor would like you to be a fanatic about their specific products, but they're all realistic as well. They recognize that they all have core strengths in their particular tool sets, and that we can't really act a sort of a marketing group for a particular vendor. Normally we deal with the vendors through their university relations people, but we also talk to former students at their current work sites to get feedback as to which tools to teach with." "As far the current prices being charged I wish all the vendors would charge the same. At a state university like Texas A&M, it's a real headache to get a license agreement from the EDA vendors that's acceptable. That's because every state university needs contracts that can be governed by their state laws and the university is part of the state government, so the contract also has to stipulate that the company can't sue the state. None of the license agreements from EDA vendors actually say that, so everybody's dancing around all of this all the time. For instance, we are still extending our current Cadence license held by the Electrical Engineering Department. I tried to get a separate license for Computer Science, but had to give up since it was going to take more than a year." ******************************* Walker's tenacity for supervising licenses and teaching notwithstanding, statistics indicate a drop-off in CS enrollment across the U.S. Walker says those statistics may be misleading. "I teach an intro course on ECE it's a new course we created last year for EE and CE students. And interestingly enough, the enrollment is that same course this year is as large as it was last year. In fact, although our total enrollment in computer engineering may be down a bit true we've had some losses as students have looked at the job market and decided to change majors the electrical engineering major is up at Texas A&M. The total overall has remained steady." "I remember when I was getting my PhD, Intel was laying people off. Clearly, as the market goes up and down, the number of relevant majors goes up and down. I think that's less so, however, in electrical engineering because that discipline also includes power engineering and power systems related to electric vehicles and so forth. Electoral engineering is a really broad field, and for many students it feels like a very safe place to spend their education. Like I said, at Texas A&M, we're just not really seeing a drop-off in electrical engineering enrollment." By the way, Walker says the situation is not changing as quickly as some would suggest in industry either. He says the outsourcing of high-tech job off-shore is happening slowly: "Based on my observations, I don't see that outsourcing means distributed design teams are on the way out all the engineering moving off-shore. The lead engineers or architects on big design projects continue to be a PhD in the company working in the U.S., Europe, Japan, Korea, or Taiwan, etc., while groups of less experienced, lesser skilled engineers in other countries are doing the implementation, and further downstream, other groups in the distributed team are doing the validation and testing." "It takes time for companies to work out this distributed design process and that's highly dependent on the company. For some companies they're still figuring out what works in a distributed fashion and what doesn't. More importantly, that process only works if the problem can be decomposed. For instance, even today I can't imagine doing that kind of partitioning for an analog circuit. So, I just don't see all of the engineering jobs exiting the U.S. for a long, long time. As far as the EDA tools are concerned, my observation is that when developers are too far away from the users, it just doesn't work. There are still lots of companies here in the U.S. hiring like mad, so I just don't think things are going to change that quickly." "Having said all of this, I'm not unaware of the competition for jobs that our graduates face, so I try hard to have all of my students intern at a company before they graduate. If a company gets to know you through that process and likes you, they'll want to hire you after you get out of school When the market was extremely tough in 2000 and 2001, the internship process got jobs for guys who otherwise might not have been hired." "Still, I maintain that the job market never got as bad as some people thought it did. You have to keep in mind that this whole H1 visa thing was about hiring people from oversees, and that included population across a pretty wide skill set. Congress has now just passed a bill for a 20,000-visa quota for people to get graduate degrees in the U.S. These restrictions will tend to put more pressure on the lesser educated the visas will go to the higher skilled instead. There will still be lots of jobs here." "Similarly, although I know that the number of foreign student applications are down everywhere in the U.S., it's my observation that it's not way down with the best students. Right now, I've got grad students from everywhere in the world. From their perspective, it's a win-win situation. They can always go back to their home countries to get a job, or they can stay here and work after they finish if their skills are specialized enough." ******************************* Walker says he's adamant about one thing above all else for his students. "I'm not denying that for American engineers, there's been a lot of bad press about lay-off and off-shore outsourcing. I have several former graduate students, for instance, now working in Bangalore where some of the jobs are. But I always tell my freshman that nothing has changed here. It's always been true and it will always be true that you've got to stay competitive in your market. You have to work hard and defend your place and your income in the food chain. It's just that now you have to think in terms of competing in the global market, not just in terms of the domestic job market." "So again, it's true that with outsourcing, there has been some decline in American students coming into our field, but the situation's just not as bad as it's made out to be. If the students at the bottom end of the spectrum who are getting hurt, which is again why I always emphasize to my undergraduates: What is your value-add compared to employees in the rest of the world? I tell them they must always be conscious of this. That's not new to engineering or any professional endeavor for that matter. Life-long learning and stay ahead of the curve will always be important." ******************************* Editor's Note: By way of setting me straight between Aggies and Longhorns, Hank Walker sent this e-mail: ******************************* Peggy, Texas A&M at College Station is the oldest public university in Texas. We are the flagship of the Texas A&M University System, which includes 9 universities and a health science center with 100k+ students. The George H. W. Bush Library is located at Texas A&M. UT Austin is the second oldest public university in Texas, and is the flagship of the UT System, which includes 9 universities and a health science center with 178k students. But before the people at Baylor jump on me, Texas A&M is only the oldest PUBLIC university in Texas. Baylor was founded in 1845. Thanks, Hank Walker ( http://www.cs.tamu.edu/people/faculty/walker)
Peggy Aycinena owns and operates EDA Confidential. She can be reached at peggy@aycinena.com
|