Digital or Analog IC for a person with digital bias

[snowman_]

Hi PF,

I can't decide which path I should go down in terms of my focus. I really love digital hardware from FPGAs to doing IC layouts, I also enjoy the trade-offs involved in making a computer architecture. I also like analog circuits with things like mosfets and opamps. I think it would be interesting to work on integrated filters, amplifiers, data converters.

Problem is most of my experience related to design (student level) is with digital hardware. I don't have much analog knowledge beyond circuit analysis, I've never tried to do an analog layout or program a reconfigurable analog system. I'm not sure why, but I feel like many students are pursuing the digital topics...I have worry that if I go with digital that I'll be a dime-a-dozen designer.

Another issue is that I'm already involved in research related to digital topics. When I go to graduate school I'll already have a few ideas for research since I've been following others work in the field. I also have connections with faculty at other universities related to work in digital topics which will help for graduate school. In terms of analog...I have nothing, my knowledge of current research is almost non-existent.

I'm applying to PhD programs (currently senior in undergraduate EE in the US) in either electrical or computer engineering this fall so I need to determine which I'm going to focus on. No matter what I choose I plan to take courses from both sides to be well rounded.

Should I go with what my current strengths are? Or attempt to slowly work towards analog topics of interest even though I don't know much beyond UG foundation knowledge. Does anyone have any recommended programs for either analog or digital electronics? I would also appreciate any recommendations into current topics in analog IC design, links to small papers would be even better.

Thank you :)

 

[JakeBrodskyPE]

The point of your education is to lay a foundation so that you can continue learning long after you have left the classroom.

If you want to learn about something else after you get out of school, study it! You shouldn't need a professor grading your work to be able to teach yourself something.

 

[snowman_]

You're right, learning is a life long process. I'm one of the few students in my classes that buys extra textbooks to read over the semester, so I do enjoy it.

I'm just nervous as some PhD programs make you choose an academic track. A lot of the lower-mid tier schools seem to lack good digital VLSI programs (at least one's I looked at) so I'm trying to see what other things I might enjoy if I don't get into a strong digital program. I really like Purdue's program but I doubt I'd get in. My current school has one VLSI course which I already took :/

 

[analogdesign]

Just wanted to chime in here really quick and stress that in practice analog design is quite a bit different from VLSI digital design (i.e. the RTL -> synthesis -> P&R -> STA/verification chain). A lot of people hate hard-core analog design (like you'd do in a PhD program) with a passion because it is really math-y, messy, and intuition based. The overlap with synthesized digital design is surprisingly small. If you're interested in high-speed logic (for example serializer design) that is essentially analog design these days but your digital bias might not be such a big deal in that sense.

 

[berkeman]

I can't decide which path I should go down in terms of my focus. I really love digital hardware from FPGAs to doing IC layouts, I also enjoy the trade-offs involved in making a computer architecture. I also like analog circuits with things like mosfets and opamps. I think it would be interesting to work on integrated filters, amplifiers, data converters.

Problem is most of my experience related to design (student level) is with digital hardware. I don't have much analog knowledge beyond circuit analysis, I've never tried to do an analog layout or program a reconfigurable analog system. I'm not sure why, but I feel like many students are pursuing the digital topics...I have worry that if I go with digital that I'll be a dime-a-dozen designer.

How much have you worked with SPICE simulations? As an analog IC designer you will be using HSPICE or a similar variation a *lot*. As @analogdesign says, the tool chains are very different between analog & digital IC design. As an analog IC designer you will likely also be doing a lot of device modeling and verification using spice -- somebody has to tune up the models for the devices that you are using as the building blocks in your analog designs.

There is another option that you might consider, to help you to distinguish yourself from the digital-only IC designers. With mixed-signal IC design (like we do at my company), you have both digital and analog blocks all working together on the same die. You could still focus on the architecture and digital toolchain, but with some extra knowledge of analog IC design, that would help you to better architect the system and floorplan the die. It also helps you to understand some of the reasons for choosing certain processes to support mixed signal IC designs -- process tradeoffs that affect the performance of both the analog & digital circuitry.

Do any of your school choices seem to offer classes in mixed signal IC design?

 

[snowman_]

Just wanted to chime in here really quick and stress that in practice analog design is quite a bit different from VLSI digital design (i.e. the RTL -> synthesis -> P&R -> STA/verification chain). A lot of people hate hard-core analog design (like you'd do in a PhD program) with a passion because it is really math-y, messy, and intuition based. The overlap with synthesized digital design is surprisingly small. If you're interested in high-speed logic (for example serializer design) that is essentially analog design these days but your digital bias might not be such a big deal in that sense.

Thanks for mentioning these points. By messy math do you mean things similar to determining your mosfet sizing in an integrated opamp?

How much have you worked with SPICE simulations? As an analog IC designer you will be using HSPICE or a similar variation a *lot*. As @analogdesign says, the tool chains are very different between analog & digital IC design. As an analog IC designer you will likely also be doing a lot of device modeling and verification using spice -- somebody has to tune up the models for the devices that you are using as the building blocks in your analog designs.

There is another option that you might consider, to help you to distinguish yourself from the digital-only IC designers. With mixed-signal IC design (like we do at my company), you have both digital and analog blocks all working together on the same die. You could still focus on the architecture and digital toolchain, but with some extra knowledge of analog IC design, that would help you to better architect the system and floorplan the die. It also helps you to understand some of the reasons for choosing certain processes to support mixed signal IC designs -- process tradeoffs that affect the performance of both the analog & digital circuitry.

Do any of your school choices seem to offer classes in mixed signal IC design?

I've used spectre for basic AOI boolean functions, muxs, different types of DFFs, and a 64k transistor LFSR. I'm supposed to get my lfsr fabricated but I haven't had a chance to connect the pads (can only use cadence at school). All simulations were done for the schematic and the extracted layout levels, we had to spend a lot of time looking at the waveforms for timing. I have 15 weeks of about 15-25 hours a week in Cadence, so even less for simulation.

For analog simulation are you looking at gain vs bandwidth..what are some key differences in the simulation stages between digital and analog?

Ohio State University has a good analog VLSI lab which does mixed-signal stuff as well, I'm applying there. I've been somewhat interested in mixed signal ever since I helped with a project that used the Cypress PSoC5, I wouldn't mind working more with it.

I'm applying to any decent program (in the US) in digital or analog VLSI, fault-tolerant digital systems.

 

[analogdesign]

Thanks for mentioning these points. By messy math do you mean things similar to determining your mosfet sizing in an integrated opamp?

Funny you should mention that because sizing circuits is one of the things you don't typically do math to figure out. It turns out that when transistor lengths are below about 0.5um or so the square-law equations for Id vs. Vgs are laughably inaccurate. So inaccurate to be useless unless you are interested in relative effects of different parameters. To see for yourself if you have good models at your school try running a simulation of Id vs. Vds for various Vgs values and compare it to the square-law graphs in most textbooks.

I've used spectre for basic AOI boolean functions, muxs, different types of DFFs, and a 64k transistor LFSR. I'm supposed to get my lfsr fabricated but I haven't had a chance to connect the pads (can only use cadence at school). All simulations were done for the schematic and the extracted layout levels, we had to spend a lot of time looking at the waveforms for timing. I have 15 weeks of about 15-25 hours a week in Cadence, so even less for simulation.

Spectre is essentially SPICE (with some weird syntax but that only matters if you're scripting it) and Cadence is the Microsoft of the IC Design world so almost everyone uses it for schematic capture and layout. You have already done some analog simulations with your logic cells so that's good.

For analog simulation are you looking at gain vs bandwidth..what are some key differences in the simulation stages between digital and analog?

Yes, those, but also many, many others. In digital you're worried about if you can get a logic level from point A to point B reliably. So, as long as you can assure you don't violate setup and hold in your design you're golden. The synthesis and place and route tools will do most of your detailed circuit work for you so you can focus on the algorithm with your RTL.

The world is different from an analog standpoint, though. There are many parameters which trade off each other so it is impossible to come up with a standard closed-form solution like you can with digital. You end up trading power for noise, or area for speed, or power for linearity, or dynamic range for speed, and on and on. It requires a lot of intuition and trying things out. This is what people love about analog design (and what some people find frustrating). Besides gain and bandwidth, you'll care about noise, distortion, dynamic range, power dissipation, stability, how it behaves across process corners and temperature, and so on.

Ohio State University has a good analog VLSI lab which does mixed-signal stuff as well, I'm applying there. I've been somewhat interested in mixed signal ever since I helped with a project that used the Cypress PSoC5, I wouldn't mind working more with it.

I'm applying to any decent program (in the US) in digital or analog VLSI, fault-tolerant digital systems.

I am personally familiar with the Ohio State program. It is excellent.

For "mathy" I meant if you get a PhD in it. I did my PhD in distortion calibration and let me tell you, that was some heavy, heavy math. Nonlinear adaptive algorithms are mind bending. I'm still in recovery. In the real world I've done a lot less math (we have tools for that) but it is still "messy".

I can give you a bit of a flavor for how analog design is "messy". In my first job I was assigned to work on a highly integrated modem chip. There were about 4 analog designers, 10 digital designers, and 20 or so software engineers (!) on the project if I remember correctly. Anyway, since I was fresh out of school I was assigned to do a temperature sensor to monitor the chip's line drivers. Easy right? I could get a design for a bandgap reference and a sigma-delta ADC out of a book. Turned out my company had a bandgap layout in the process we were using already to go. I had the circuit working in less than a month.

Then the fun began. The guy designing the clock synthesizer asked me what jitter I needed on the sigma-delta clock. What? I had no idea. Did the sigma-delta noise-shaping loop filter out jitter in the sampling clock? I didn't know, and my textbooks and journal articles were silent on the matter. So I had to do some math to figure out how to simulate that and finally felt comfortable with the answer. Then the digital team gave me some models about the cross-talk I could expect. Guess what? It totally hosed my ADC. Back to the drawing board. Turned out I needed to re-do a lot of the layout.

Then the software team asked for a bit of programmability in the temp sensor. That required some new circuits and rethinking of the input stage of the ADC.

Then the packaging team got a hold of me and asked about my current requirements. I got a model of the package and simulated it with my ADC. Disaster! The power supply bounced all over the place. I hadn't seen this before since I had underestimated the package lead inductance in my simulations. My decoupling capacitors were resonating with the package. Now what do I do? The textbook doesn't even talk about no-chip decoupling.

So, I had to do some more math. I made an RLC model of the system and discovered that the resonance frequency was dangerously close to one of the clock harmonics. How did I fix it? Well I added some series R to "de-Q" my capacitors and damped out the resonance. This type of thing went on and on. When the chip leads did top level integration of the whole system I had a new list of to-dos. It took me six months and a lot of late nights to get that "easy" temp sensor out the door.

So that's a taste of analog IC design. If you like circuits it's the kind of job where you'll pinch yourself that they are actually paying you quite well to do it (I am amazed every morning). It's not for everyone but if you think of yourself as someone who solves "analog problems" rather than someone who designs "analog circuits" you'll do well.

That's a lot of writing, but when I was at your level someone gave me a breakdown of what life was like in different roles in the electronics industry and I'm glad he did. So I'm just paying it forward.

 

[snowman_]

Thanks for the run down, it gives me a lot to think about. The experiences you've described sound very interesting.

I'll edit in more as my head digests the information.

Would've replied sooner but it was final exam week.

 

[Phil Xu]

<< Mentor Note -- spam post deleted >>

 

[snowman_]

Saw old thread pop up. Ended up getting RA offer for emerging device group, will be doing spintronic related stuff. Thanks for the advice last year.

 

[analogdesign]

That's great, snowman_! I hope you have a fruitful time in graduate school and succeed in your future endeavors. We don't normally get to hear what happened so thanks for taking the time. It makes me smile to know it worked out for you.