Taking students from schematic to silicon in one semester (2018) [pdf] | Hacker News Reader

Taking students from schematic to silicon in one semester (2018) [pdf]

people.eecs.berkeley.edu

146 pointsteleforce24 comments

I studied Engineering Physics and specialized in semiconductor engineering. In our courses we made a working SAW-filter. We did everything from design, to silicon fabrication,to bonding, packaging and testing. It was a fun project, I learned a lot and I still have my little SAW-filter.

To add a little background, I think there are two main reasons that made this project possible:

1. Our professor was very good at organizing donations of decommissioned equipment from the industry. The department was well funded and that certainly must have helped a lot, but what made all the difference is that our professor was much better at getting in-kind donations than most of his colleagues.

2. Selecting the SAW-filter as a project was a brilliant idea. It enabled us to make a completely working product while teaching us basically most essential steps that are used in the production of real active semiconductor components too. At the same time it allowed us to skip the really hard, expensive or dangerous tasks like high-resolution masks, multiple masks and the whole business of doping.

What school did you go to? Engineering physics sounds right up my alley and I didn’t know it was a thing

There should be another version of nand2tetris called sand2nand

Hide buffer time in the schedule: we were able to manufacture a chip in large part because we never discussed the foundry’s true due date with students. Instead we opted to build tolerance into the schedule and kept students aiming for an earlier deadline.

Having worked in the industry (thankfully not anymore), surprisingly that's how it works in the real world as well. Engineers are given a tape-out date that they have to meet, but managers have some hidden extra slack just in case.

"The tendency to spend valuable hours practicing perfectionism resulted in weeks of timeline delays and loss of design integration and verification time."

Students in a nutshell.

Anecdotally, the "II. Overcoming Student Misconceptions" section really rang true to how I've experienced working with highly academic engineers - ie: wanting to put blinders on only working on their component, late is assumed OK, not anticipating + coordinating with other teams interfaces/integration boundaries, not designing with real-world implementation in mind.

Stanford had (has?) a class where you actually fabricate the chips in their cleanroom. I've heard from people who took the class that you'd have to reserve individual pieces of fabrication equipment at all sorts of weird hours (not unlike a real chip plant) and work shifts to get done in time

The University of Michigan has a class where we came up with an idea for + designed a 2mm x 2mm analog/digital chip and a MEMS chip, and got to go into the school's nanofabrication facility to help manufacture the chips. Everyone got to take one home. It was the most memorable class of my undergraduate career.

In college, I had the privilege of taking and then teaching a VLSI class like this. It was an amazing experience and I think more engineers need to go through the process. When I get the chance (no IP agreement with a company), I want to do the efabless thing that Google is offering

The VLSI I took for my EE degree was the hardest course by far. I did a digital logic project and the amount of time I spent looking at my traces and trying to figure out what went wrong, only to forget to ground my P-wells or make one of transistors too small. Maybe I just wasn't good at it but have a lot of respect for people who do it professionally.

There's also MEMS which is basically making sensors (even motors) out of Silicon.

This would be a really great experience but only for a small subset of EEs who make it into RF/Analog IC design. And even so, I'd much rather see strong fundamentals than tape out experience in a new college grad because the latter is much easier to teach than the former. In fact, with no prior tape out experience (but an MSEE), I was responsible for a couple of tape outs after only a year of experience at my first IC design job.

I'm surprised that this course was only offered first in 2017 and isn't a degree requirement. I'm not an electrical engineer, so I don't know if these skills are taught in a different and less hands-on manner, but it seems like being able to work in a team with tight deadlines and being able to design a chip for manufacturing are key skills that an engineer should have.

IC design/VLSI classes where you do this are offered at most decent universities with an EE department. These classes are a big part of why MOSIS was founded, which has been around for about 40 years now.

I'm thinking probably less than 25% of EEs take this track though, because there is a lot more to EE than just IC design.

Yeah, I took a similar course around 2010. It was a fourth year course and had a ton of prerequisites, so you pretty much had to plan to take it from second year.

Most EE students I knew decided to specialize in power generation since it was seen as the easiest path to a cushy job.

There's generally a required class that involves implementing a chip design on an FPGA.

A very small fraction of people getting EE degrees actually go into IC design and it would probably be very expensive/a logistical nightmare to put such a large number of students through this type of class.

We had a course like this at my engineering school. Cutting your own chip is a fantastic journey. My partner made a key mistake along the way by reversing the layout of an op amp, but I was able to model his mistake and show the prof that I had figured it out on the lab bench. Full marks. It was fun.

I would love to do such a project in my free time, but I can't afford the needed software tools.

Google and efabless are offering free IC production for people who want to do open source chip design for what it's worth: https://www.efabless.com/open_shuttle_program

This started last summer and a number of interesting designs have been produced, pretty much entirely with free, open source tools.

Join the slack and ask around for projects that need help if you want to dip your toes in this sort of work! https://invite.skywater.tools/

Is it still digital only?

I'm interested, but almost entirely on the analog side -- I'd love to design a NLTL, VCO, or distributed amplifier. I have the utmost respect for the crowd looking to scale RISC-V cores, but it's just not my scene.

EDIT: I took a look at https://github.com/yrrapt/caravel_amsat_txrx_ic_mpw2 and there are clearly some analog projects in there. Evidently I should have hustled harder!

No, they have a harness called Caravan that has no I/O buffers. I calculated Ft=65 GHz on this process and have a 1-15 GHz amp working in simulation (pre-extraction), though I don’t know if I’ll have time to apply for the next run.

From reading the documentation, it appears that their open source tools start with Verilog RTL files that have been simulated with Synopsys VCS, Mentor Graphics Questa, or Cadence Incisive Enterprise Simulator.

That said, one could probably use the free (but not open source) Vivado HL WebPACK to create and test the RTL design and then follow along with the Skywater tools. But that way, the way towards the FPGA would still be closed source and only the path from FPGA on to ASIC would be open.

Further down the rabbit hole, I read about https://github.com/icebreaker-fpga/icebreaker which appears to be a tiny Teensy-compatible open source FPGA board with accompanying open source tools. Yay!

Thank you very much for bringing all of this to my attention :)

I tried messaging instructors at some Berkeley course and asked if I could watch the videos (didn't get a response).

I also followed the Google/efabless program but it doesn't come with any handholding (kudos to the key google engineer who organized it .. he seemed very passionate from the talks/meetups I attended).

Ultimately, I got demotivated from my personal quest to fabricate something from scratch. I think FPGAs are the sweet spot .. they let you see what is going on without any of the physics.

Good to luck to you on your quest.

Tools are free if you know where to look... ;-)

But the fab costs are still huge, especially if you don't get it right the first time.

Reply on news.ycombinator.com