Question from Sidharth : Course Selection, MS vs PhD, Future scope for VLSI Design

I'll try to address the question Sidharth brought up in the comments for the last post. So first a quick note on course selection at American universities.

Mechanics of Course Selection
In grad schools in the US, you have almost an unlimited choice of courses to take. You could take courses in virtually any field within your department and outside. A lot of students try to learn how to play tennis, golf or ball-room dancing while in grad school, although these courses don't get you any closer to getting a degree. So a lot of students take one such course every semester. One constraint for course selection is that if you are working with a professor for your research, you may be required to take the course your professor is teaching that semester.

At UW's ECE Department students frequently took courses in Computer Science. For instance if you are trying to learn Computer Architecture, you might want to take courses in Operating Systems or Compilers in the CS Department. You are the boss. You decide how you want to complete your course requirements. You could ask professors for course advice too, but you make the final decision on course selection. So in Sidharth's case, you could take courses in VLSI Design and Networking as well.

Deciding what you want to do
Before I get to the question of future scope, an important question you need to answer is whether you want to do an MS or PhD. Honest advice if you want to do a PhD, do it right away. It gets much harder if you leave with an MS, to return for a PhD. If you want to head to a higher ranked university, apply in the second year of your Master's so that you can leave at the end of your MS.

A second question you should consider is are you interested in research or in a development job? Although semiconductor chip makers do very high-tech work, not all of it can be classified as research. (in fact most of it is not research). My definition of research is the pursuit of new knowledge, new truths. The intention is only to prototype and publish some new techniques or inventions, but not really to make a product out of them. Development engineering starts out at these ideas that came out of research and creating products out of them. While there is a lot of thought needed in development jobs and there are challenging tasks, there is also a lot of drudgery. If you are looking to become a researcher, you need a PhD in nearly every case because these days 33% of engineers have Master's degrees in the US. (Source EDN Survey 2008) Also the field you choose for your research has a big impact on your future. Entering nascent fields like nanotechnology, quantum electronics, bio computation, EUV lithography, natural language processing could ensure that you do get to do research eventually. VLSI Design as a broad field does include a lot of different areas, but they have been around for a couple of decades now. So it is a good field to be in if you plan to get a development engineering job.

The scope for semiconductors : VLSI Design
We all hear about how computing has changed our world and how this is just the beginning. There are a lot of interesting problems that still need to be solved. Networking all the devices in your home, creating faster computers, 3D stereo vision displays, personal robots, the list is long and the only limit is your imagination. A lot of prominent experts are sounding cautionary notes about the future of silicon CMOS technology though. There have been a lot of reports in EE Times about challenges to continued integration and there are some indications that creating transistors at 32nm feature sizes and beyond might be very expensive. This might prevent a lot of companies from creating chips if the costs become prohibitive. I am no expert in how this will all work out but there are varying predictions for the future of semiconductor scaling for 5 year or 10 years. Shekhar Borkar of Intel claims that there will be a lot of problems, higher degree of variability, higher failure rates, higher transient error rates and how we might have to add a lot of redundancy to the logic on chips to correct the errors. Short answer is that until some big breakthroughs are made the future doesn't look without risk. We could be optimistic and assume that some smart scientists will be able to solve the question of lithography and create an EUV technology that works, find ways to minimize error rates, but things definitely won't be as easy as the last 20 odd years in the chip business.