Tác giả giữ toàn bộ bản quyền (All Rights Reserved) — không phải giấy phép mở. Bạn vẫn xem và học miễn phí trên OpenER; muốn tái sử dụng cần xin phép tác giả.
DAN GARCIA: --session.
BORA NIKOLIC
Okay.
DAN GARCIA
And we'll start it.
BORA NIKOLIC
Okay, we're on.
DAN GARCIA
We're starting. Perfect, we're live.
BORA NIKOLIC
And people are here.
DAN GARCIA
People are here. Welcome, everybody.
All right, good to see everybody here. All right, let's get started.
Ladies and gentlemen, welcome to week 7, once a week lecture, 61C, Logic, RISC-V datapath. [CHEERS] Great to see you.
BORA NIKOLIC
We are almost at the halfway point. Can you believe that?
DAN GARCIA
That's amazing. It's like doing a podcast with you, Bora. It's really great. It's like a video podcast, where we go back and forth.
BORA NIKOLIC
Yeah, a radio show — radio show, yeah.
DAN GARCIA
It's like a radio show. I love this. This is great. It's great to see everybody, as always.
We've got a full agenda. We should just jump right in.
BORA NIKOLIC
Yes, exactly.
DAN GARCIA
So here is our first — yesterday's agenda. Again, welcome to week 7. And as always, we do the same script we did before. We have computing in the news.
We're going to mention the fact in a moment. We have our wonderful guest, Victoria Coleman, who's going to join us, and Bora will lead that conversation. We do the week's plan.
And then we actually didn't add this to this, but stay tuned, because we're going to review the workload and wellness survey.
So we didn't add that to the agenda, but that is on the agenda today — to have a deeper review and then a conversation of what do we take out of that?
So we're have a nice conversation at the end of it after we bid Victoria adieu after her introduction. So stay tuned. This is going to be an exciting time.
No big announcements, but a nice conversation hopefully around the output of that workload and wellness survey. Big tech news — Bora, take it away.
BORA NIKOLIC
Yeah, we mentioned this already. I mean, there is a huge bill that is going through the Congress and the Senate about semiconductor industry and generally design in the US.
It is depending on who you ask, and which ask you ask somebody, it's up to $22 billion over the next 10 years. That's going to change a lot of things.
And it is going to change the perhaps trajectory of your careers, you who are going to be entering this field in about two years.
There is a lot of stuff that is going to happen there.
So we do have our guest today that we tried to get a few weeks ago, and that is Dr. Victoria Coleman.
We actually tried to get her here on her first day when she became — when she was sworn in as director of DARPA, which is Defense Advanced Research Projects Agency.
But that was a bit too tough. So she's here two weeks into her new job.
You guys need to appreciate how big of a deal is to have a DARPA director show up in a class. And we are very, very, very appreciative about this.
So we're going to ask Victoria a few questions. And hopefully, her audio is doing better now. Hi, Victoria. Thank you for joining us.
VICTORIA COLEMAN
You're welcome. Can you hear me at all?
BORA NIKOLIC
Perfect. Perfect.
DAN GARCIA
Thank you. Thank you. Woo! We got it solved.
VICTORIA COLEMAN
I figured out how to pair my pods today to my MacBook. It's wonderful. Technology is amazing when it works. Thank you, Bora, and thank you, Dan, for inviting me along.
And I have to start with an apology. I did try — well, I did try and fail to join you literally the first day on the job. Time got away from me.
But I'm really pleased to be here today.
As some of you probably know, I have a weird little appointment at Berkeley that I'm very proud of.
And I was really pleased, actually, to be able to keep my appointment while I'm at the agency. So I'm really happy. So when this tour is over, I'll be back on campus.
And we'll have many more war stories to exchange when I get there. But for now, I'm out here on the East Coast, missing California, but really pleased to be here.
And if I can offer any thoughts, any insights into the kinds of things that you had in mind, I'm happy to. Bora, you talked a little bit about the legislation.
Is this something you'd like to talk about?
BORA NIKOLIC
Let me ask you a few — VICTORIA COLEMAN: Yeah, go ahead.
BORA NIKOLIC
A few questions first.
So this is perhaps the largest computer architecture class in the US and likely in the world at the moment.
DAN GARCIA
And maybe in history — maybe in history also.
VICTORIA COLEMAN
That's great.
BORA NIKOLIC
Many of us are still learning over here how to handle that.
We decided to keep students animated and entertained by bringing interesting speakers every week. This is something that we couldn't do.
I mean, nobody would fly from East Coast or from somewhere else to come to a class. We usually ask our speakers to introduce themselves.
I know that you had a very long and productive career in industry. And we know that you're sponsoring research at Berkeley from different positions that you were in.
But what we would like to — briefly, for our students, tell us, what do you do? And how did you get there? And then I'll ask you a follow-up after that.
How does one become a DARPA director?
VICTORIA COLEMAN
All right, so a long and winding road and, frankly, a sequence of unlikely events.
As you can tell by the accent, I wasn't born here. I was born in Greece. So age 18, I packed my bags. I went to the UK to go to university there.
I'm a computer scientist. So I'm — a lot of sympathy for a class like this. I mean, I'm really excited to hear this is the largest computer architecture class.
I'm glad that somebody still working on architecture. It's wonderful to hear.
I think, honestly, it's — well, two things.
One is it's about what you want to do, and the other one is about the environment in which you work, giving you the opportunities to do these things.
So when I first came to the US in 1998, I worked at SRI, which back in the day used to be the Stanford Research Institute.
So we had a lot of work going on with DARPA. In fact, all the work that I did was with DARPA. And I fell in love with the agency.
And I had become a computer scientist because I watched — I'm old enough to have watched the landing on the moon. It was the first thing that we watched on the TV.
I remember my dad went out and bought this grainy black and white TV set. So that changed my life.
And it was interesting when I realized that had it not been for DARPA, we would have never gotten to the moon, because Saturn V was a key component of the Apollo missions.
I just fell in love with it. I left SRI in 2004, because as an academic — some of you will appreciate this.
As an academic, you aspire to get a real job, a so-called "real job," so getting closer and closer to product.
So I spent a dozen years working in corporate in companies like Intel and Samsung and HP.
But I always wanted that — I knew that going to the agency or working in S&T for national security was the place that I was most excited about.
So I did a lot of volunteer advisory work as I was doing — as I was pursuing my career in corporate.
And it came to the point, frankly, for me, personally, that instead of telling people that I was going to work to support my DOD habit, I ended up working 130 days a year on top of my day job doing the DOD work.
And then it dawned on me that if I enjoy this so much, maybe I should do it for a living. So that was the big light bulb. And I decided to join the agency.
And now it's kind of interesting. You ask how does one get to be DARPA director? Being DARPA director is an exceptionally amazing job.
I think for me, in my book, it's the best job on the planet. But there are many other very, very good jobs at DARPA.
So the program managers at DARPA have a ton of fun. Actually, I think they have better — their job is a better job than the director job. My appointment is political.
But many people don't know this, but the director of DARPA is the — it's a political appointment. It's the only political appointment in the whole of the agency.
So you become the DARPA director by virtue of the administration. And of course, every administration has different ways of picking its people.
In this case, I think it was a matter of respect for the work that I've done in the past, and also hopefully some trust about the work that I will do in the future in the agency.
It was really quite emotional, really. I was sworn in the first week I was here. And I was sworn in by Michael Kratsios, who is the CTO of the United States.
And he also is the acting undersecretary of defense for research and engineering. And he is also of Greek descent. So I was thinking, what are the chances of this?
Two Greeks, two pseudo-Greeks, in the chain of command — DARPA and the department. I don't think there are many places in the world where that can happen, honestly.
Yeah, I feel fortunate. I feel humble. Actually, week two, I'm a little less worried. Week one was like, let me not break anything.
This is an amazing, amazing agency. And I mean, of course, there's nothing that is so perfect that we cannot make it better. So that's what we're going to try and do.
But first of all, we're not going to break anything. Bora, is this close to what you were trying to elicit?
BORA NIKOLIC
Yes, thank you very much. That's great. I mean, students probably don't know that you don't have very many layers of reporting above you.
This is just a bit of a joke. Students may not know what Siri is. And probably the best widely known product of SRI is Siri.
So students don't know what SRI is, but Siri is the best-known product of SRI. And we don't mention what does it stand for.
It has a word that we don't mention over here in Berkeley. It's a Stanford research institute — has nothing to do with the Stanford campus.
We don't do that.
Quick, another quick question. So this CHIPS Bill has been going on for a while.
And every time I try to read something about it, it changes shape, which is how things happen when they're propagating from — through the government.
It has one of the things that we have recently seen through the announcements — that it has a big push for the workforce.
Tell us something about the CHIPS Bill and this part of it that has to deal with the workforce.
Because we have 1,200 people in the class, and this may actually affect what are they going to be doing for the next 30 years or so.
VICTORIA COLEMAN
Yeah, it's kind of interesting. I mean, when we talk about semiconductor, and the first thing that people think about is manufacturing. It's Moore's law, blah, blah.
But the end of the day, as you say, it's all about the people. And it's kind of interesting.
Back in 2018, the Defense Science Board did a study on microelectronics leadership in the US. And there were a number of recommendations that were made.
One of the recommendations was to vastly increase the number of sponsored PhD positions that the department has with particular emphasis on microelectronics.
And the idea there was that the department would pay for people's PhDs, and then people would give back by committing to work the same number of years that they were sponsored by the department into either a government lab or a US-based company.
So I think that is part of the picture.
But of course, you don't get to do a PhD in semiconductors and microelectronics if you haven't started probably at age 14 or 15, directing yourself in that career.
So it starts much earlier than that.
It turns out that actually, the department has a program. It's a small program right now that starts by incentivizing high schoolers to consider microelectronics as a career.
And that helps them through undergraduate years and certainly through the graduate program. So I think these are some of the tools that we have at our disposal.
But I think there is — behind your question, there's a bigger strategy, right?
And for a long time in the business, there was this perception that it didn't matter if semiconductor manufacturing went overseas, because we kept the "high-value" part of the process here in the United States — namely, design and architecture and EDA
and so on. So for a long time, people weren't paying attention to the talent equation of this.
But one of the things that slowly becomes true if you don't pay attention is that you wake up one day, and you don't have the skilled workforce domestically to even do some of the innovative work on the design and the EDA.
Because think about that. If you're an aspiring PhD student, you're going to go and spend five, six years of your life, say, at a school like Berkeley, at a school like at MIT.
You will only do that if there are good, great jobs to be had here in the United States.
And if there aren't, then you're going to take your skills and your talents and go to the places that do have those opportunities.
So I think gradually, what you're seeing is a deskilling of the workforce here domestically if you let manufacturing go overseas completely.
I had the great fortune of working a little bit with Andy Grove when I was at Intel.
And he really believed that, unless you were able to scale manufacturing, semiconductor manufacturing in the US, you will eventually lose the plot.
And he's somebody that really ought to know, right?
I think him and Bob Noyce single-handedly crafted the semiconductor business that we know today — and of course, another glorious alum of Berkeley.
So I always felt that his words resonated.
Building enough scaling capacity here at home to attract the best and brightest and keep them in the fold so that we could indeed have them be part of the innovation army, if you like, that creates the amazing ideas.
But I don't think this industry is all about ideas. It's also about making things. And we need to have both here domestically, which circles back, Bora, to the bill.
So the CHIPS Bill, and before that, the American Foundries Act — they're all trying to get at this problem. I mean, everybody is watching the rate at which manufacturing of semiconductors has shifted overseas.
So I think today, we have 12% of global production steady state. I believe in a decade, we'll be at 9%. So that's a very uncomfortable kind of place to be in.
So there are lots of people on the Hill, in companies, in the department, in the associations, s trying to figure out how can — diagnose, first of all, why this is happening, and then to try and find some ways of problem-solving.
Part of it clearly is, for the lack of a better word, incentives.
I mean, one of the reasons why companies like TSMC are very successful, as well as having great technology, is that they get a lot of support domestically from the government.
Here in the United States at the federal level, there are really no incentives for semiconductor companies. So you may agree or disagree with that.
But what it does do is it changes the economics of manufacturing here versus in a place where these incentives exist.
So many folks are trying to get their head around what would happen if we were to inject some incentives in the business. Would that change there? Would it change the economics enough?
And would it change the economics long-term for us to be able to redress the balance, if you like? So that's part of the conversation on the Hill.
The other part of the conversation is what, does the DOD need? Because we certainly need to have access to high-end parts.
And if, for example, all the production that we rely on is overseas, perhaps the supply chain might dry up for a variety of reasons. So the DOD has very specific needs.
So this bill's — as well as looking overall at industry health, they also are looking out for the needs of the department.
And also, everybody comprehends that the way that we lead is by investing in R&D. So if you were to abstract away from the detail, there are these three big pieces.
There's what are you doing to redress the financial balance, the economics? What do you do for the Department of Defense? And then what do you do with R&D?
And of course, the R&D piece is where the workforce piece falls in squarely. That's where one would look at.
In fact, I think the latest version of the bill calls out specifically workforce development as one of the four of key tenets of what they call the NSTC, the National Semiconductor Technology Center.
So I think there is recognition. But of course, we need to keep reminding ourselves that it's — without the workforce we have nothing.
And it's complicated because it's not just about the people that wear the bunny suits in the fab.
It's also about building a pipeline of talent that will supply the talent that we need in all the different steps of the lifecycle of semiconductors, all the way from inception, fabrics, to packaging and delivery.
BORA NIKOLIC
Thank you. We have a couple of questions from students. We need to keep it short because we have other things to do in this class, but a few things.
So most of the students in this class will be in the design pipeline, meaning there is very little use for the fabs if there is no — if there are no chips to be manufactured.
VICTORIA COLEMAN
That's right.
BORA NIKOLIC
That we need. And in about three weeks, we'll have d another 400 people who have designed microprocessors from the gates up.
And that's quite exciting. And that enables a path, if they want to pursue, to go that way.
One thing that we have noted is this class is actually sponsored by Apple because they've perceived the same issue — that there is not enough designers around here.
And they prefer to hire locally. We got two quick questions from the students. One is, what is approximately DARPA's annual budget? And where does it go, to companies or universities?
VICTORIA COLEMAN
Okay, so the budget of DARPA varies from year to year. But roughly, it's about $3.5 billion annually. I could not tell you exactly what the breakdown is.
But DARPA typically sponsors research at universities, at companies, large or small — and small. DARPA does none of the work inside the agency.
The way that we work is by bringing on board program managers who are experts in their fields. And they come with a strong desire to solve a problem in their domain.
They pitch a program. If they're successful, they get money.
And if you were to squint and look at these programs, you'd say the other side of the program would be maybe $40 million over, say, three years.
This is roughly the level of effort that they typically walk away with.
The other thing that you should know about DARPA — it is an agency, which means it has a great deal of built-in independence in picking the elements of work that we do because we do very high-risk work — in other words, work that honestly should not be relied upon as a source of roadmap elements, because high-risk work sometimes doesn't work out.
But on the other hand, when it pays off, it really pays off big time.
So DARPA and the legislation around it is crafted in such a way as to give the agency the freedom that it needs in order to really innovate out of the box and do amazing things.
And also, it has the funds that it needs in order to be able to build out these ideas.
And because it has — because we're a part of the defense ecosystem, we also have pull from our customers within the department.
Having said that, as you know, sometimes many of the technologies that we sponsor end up being dual-use technologies. So they end up in many other places as well as the armed services.
BORA NIKOLIC
Our students may not know that RISC-I and RISC-II were DARPA projects back in the '80s with Dave Patterson because DARPA was looking for a disruption.
RISC-V was not developed by DARPA, but the first prototypes of RISC-V processors were used in DARPA projects.
One final quick thing, the question from a student, is now that you are at DARPA, are there any specific domains or technologies that you'll be trying to promote?
VICTORIA COLEMAN
Microelectronics. [LAUGHTER] BORA NIKOLIC: All right, that's great. Thank you.
VICTORIA COLEMAN
I think we are already thinking what we should be doing beyond ERIs.
So many of you will be familiar, I think, with the Electronics Resurgence Initiative, which is a program that DARPA — actually, it's a collection of programs that DARPA put together starting two years ago, the idea really being to reverse this trend of leadership in microelectronics sliding away from our hands here
in the United States.
The program was very successful in engaging non-traditional players in the research that we do — in particular, companies that typically would not be defense contractors — the Intels of the world, for example.
We are already looking to see what the program should look like for the next five years.
And we're also looking to really problem-solve around some structural aspects of manufacturing. Again, it's all about economics.
The reason why manufacturing has gone abroad is about the economics.
So oftentimes, DARPA is looking for work that has these levers — that we apply a little force, but you have a big effect.
So we are pulling ideas together internally, and in 30 days get — so don't hold me down some of this.
But you can expect to see new ideas and new interest at DARPA for building out what we do in microelectronics, but also how we connect innovation to microelectronics with what happens in the world of computation.
Because it's not just about the hardware. It's not just about the software.
We'll also be looking at ways of connecting the two so that any innovation that we put forward on the hardware side actually manifests itself and provides value to the end users, who are very unlikely to be just hardware users, right?
That's not how typically people use microelectronics. Like I say, I probably should have said check back in a month or two. But these are my thoughts two weeks into the agency.
And the way that the agency works, of course, is that there are lots of very, very smart people that have great ideas.
One of the things that I'm working on is bringing them all together and having a good, old brainstorming session and deciding how we should prioritize the kinds of things that we all want to do.
And what should we do more of and less of? So stay tuned. In the next few weeks, months, these things will become clear. And I'm from the Bay Area.
My home is in the Bay Area. I'm going to be out there a lot.
So I hope to be able to come by and perhaps meet you all in person when COVID-19 permits the opening of the campus. And by the way — BORA NIKOLIC: Thank you very much.
VICTORIA COLEMAN
I'll just put a plug into that. DARPA is doing a ton of work on COVID-19, so stay tuned on that as well.
BORA NIKOLIC
Thank you very much. These folks in — the students in the class are undergraduate students, almost all of them.
And I think they will only in about a couple of years realize, oh, we were in a Zoom room with a DARPA director.
DAN GARCIA
That's exactly right. They don't appreciate it at all. [LAUGHTER] VICTORIA COLEMAN: --keeps you grounded.
BORA NIKOLIC
Thank you very much. Thank you for — DAN GARCIA: Thank you so much for taking the time. Wonderful.
VICTORIA COLEMAN
You're very welcome. Enjoy your class.
BORA NIKOLIC
Thank you. Thank you.
VICTORIA COLEMAN
I'm very jealous. I wish I could take the class with me. Maybe you all will apply for DARPA.
BORA NIKOLIC
Oh, I mean, Dan was on NBC News, just so — [INTERPOSING VOICES] DAN GARCIA: Very sweet.
BORA NIKOLIC
Thank you. Thank you.
DAN GARCIA
Good to see you. Good to see you, Victoria. Wonderful.
VICTORIA COLEMAN
Thank you.
DAN GARCIA
That was so great.
BORA NIKOLIC
Thank you.
DAN GARCIA
Wonderful. That was great.
BORA NIKOLIC
Okay, back to — DAN GARCIA: It doesn't get much bigger than that. People don't realize what we're getting. We're getting CEOs, DARPA director. Do you realize who we're getting this semester?
What an amazing year we're having.
BORA NIKOLIC
And there are only two levels of reporting above her, just so you know, guys. I mean it's spread pretty thin.
DAN GARCIA
Pretty remarkable. Pretty remarkable.
BORA NIKOLIC
Yeah, okay.
DAN GARCIA
We're going to jump right into the schedule, yeah. So let's take a look at that. We're in the week where we're looking at finishing up logic blocks. That's my last lecture this week.
And then we jump into datapath and control. These are among the most exciting lectures that we have all year. In fact, I want to celebrate.
Bora, we'll have a special something next week when we finally make the connection with control. We're finally going to be able to make the hardware and software link.
And we'll have some celebration live with all the students. That's really going to be fun. We're finally coming together. You saw this whole printed material starting Monday, 9/28.
We're biting that burger, a veggie burger, if you'd like, all the way from the top down at the software angle. You now have machine code, and now we're biting it from the bottom up.
So SDS was the lowest level, then state is higher than that. Logic and logic blocks are above that. Then you have these bigger blocks, bigger logical blocks, that you can now wire together.
And part of what the wiring together is what's called the datapath. And then you have to then control them. You have to control the muxes.
If you've watched the lecture on logic, you'll see — and actually, it's today's lecture on logic blocks talking about what a mux is. How do you wire those muxes?
Muxes are like 2 to 1 guys, or maybe 4 to 1 guys or 8 to 1 guys. How do you control these guys? Who's the marionette that does it?
That's the control. So the brawn is the datapath, and the brain is the control. How do you even set that up?
So all of that is what Bora is going to be sharing through his lectures Wednesday, Friday, and then next Monday. And then we jump into how do we make that faster?
So how do we even jump in? By the way, all this material that you see there on this slide is on the midterm.
The part that's not, though — Wednesday, 10/21, and Friday, 10/23, is not. So he's not even listing it here. This is why we're doing that. So we're listing all the material that's there.
Pipeline means about how do you make that faster if you had — and we'll make analogies to laundry and all those things. So that's gonna be a great set of lectures for that.
And then we talk about caches. But you didn't hear me say that, but that's what we're seeing after that. And that's all post-midterm material. But caches is right after that.
So looking at the deadlines for our project — I'll get myself out of the way of this slide here. You can see.
So looking at the deadlines here — project 2b is due today, 10/5. Lab 5 is this week.
I think we had a break before for the lab, and this is talking about Logisim to get you set up for project 3. That's critical.
You got to learn how to use Logisim to be able to work with project 3, and you can see here. And there's a connection between the colors. [INTERPOSING VOICES] DAN GARCIA: Go ahead, Bora.
BORA NIKOLIC
One quick comment — lab is pretty light. I've done it. The previous version of that is pretty light. It shouldn't be very time-consuming.
But it is to introduce you to Logisim, which will enable you to do the project and I think the next lab as well. Logisim is a bit finicky tool.
You need to be disciplined and pay attention to that lab. There are many gotchas that you could get into if you are not very disciplined.
So pay attention to that. Go through it. It's not conceptually challenging. It is just learning the tool.
And then when you learn the tool, you can build anything, including a RISC-V pipeline.
DAN GARCIA
And it's remarkable. I mean, when Bora and I went to school, they didn't have these tools. We were all doing this on paper. The fact that you have these simulators is just remarkable.
So the idea of Venus and Logisim changes the entire education experience. So we're really appreciative of the folks who at Logisim who built those tools that we use.
BORA NIKOLIC
Speaking of that, I think we have Ryan and Ryan. I think one Ryan is going to give us a five-minute preview of the project.
DAN GARCIA
Oh, I love it. BORA NIKOLIC: --ties into the lecture. So if you watch this Wednesday's and Friday's lectures — and Wednesday's one is up, and Friday's should be.
I put all the videos where they're supposed to be. They just need to be linked together. So as soon as you watch that, you can actually — we will launch the project tomorrow.
And you can take a look at it and get going with it. It basically follows the project, but doing it for real in Logisim. So, Ryan, take it away you've got about five minutes.
DAN GARCIA
Welcome, Ryan. Thank you so much. Yup, I'm going to pin you here. Here we go — add your spotlight.
It's all you, my friend.
RYAN
All right, so, hey, I'm Ryan. I'm a master's student. Bora is actually my advisor for my master's degree. And I'm the project lead for project 3.
And then the other Ryan, Ryan Thornton, is the co-project lead.
So project 3, as you heard, is basically taking this simulation tool, Logisim, which is you basically draw out logical circuits — so wires, and gates, or gates, more complex things like adders.
And you're going to create the RISC-V datapath and eventually the control logic tree in the overall core, like you've been learning about in lecture.
So the key component of this project is just having a good fundamental understanding of the RISC-V instruction set and how each instruction actually works.
I think the lectures from this week, which go over the datapath in particular, and the lab, which was for Logisim, together will give you that foundation from which you can build all of this.
So some quick updates, tips, and tricks and things to be aware of — the first is you're going to get this Logisim out of the Logisim binary JAR file out of the project starter.
Please, only use that one. It's a custom one that I've added some modifications for that help with this class.
And if you use the wrong version, it can lead to some complications down the line. The second one is — BORA NIKOLIC: I did that already in my attempt today.
It causes — you don't want those.
RYAN
It can get messy.
BORA NIKOLIC
Yeah. I thought they had an installation that looked pretty good. Do you need anything to share? Are you able to share?
Is there anything to share on screen?
DAN GARCIA
Yeah, is there a demo? Are we gonna see a demo here?
RYAN
I'm gonna release — DAN GARCIA: You're good.
RYAN
Yeah. I'm so I'm going to release some demo videos that I'll record probably later tonight.
And they'll be in the spec when it gets released, of me walking through how to do one of the components, a little toy example, so that everyone can see the way that I, as a TA, think about this, and how I'm building components and how I'm testing my work.
So hopefully, that will help you all as you're working on your project understand better the flow that is designed for you. So just another tip or trick is to be very careful.
Read the spec first, all the way through, and then start. Spec reading is a very important part of the design process. And just make sure you understand everything.
Feel free to ask questions on Piazza if you need to clarify anything. It's better to understand that first and do it second rather than do it first and ask questions later.
And then the third thing is that we have a design contest this year, thanks to being sponsored by Apple.
So at the end of this contest, we're going to take all the submissions that pass 100% of the tests hidden and visible. And we're going to convert them into gate-equivalent models, basically.
So we're going to say you had this adder which is equal to 200 NAND gates.
And then we sum together all the components of your circuit, and we get this one composite score, which is basically how big is this thing you've made.
And then the lowest size will basically win points for this section of the contest. So another thing to think about as you're designing is just, what's the simplest way I can do this?
You can do things a very complicated way using very large blocks, or you can abstract it down to the simplest logical forms and use small blocks and small wires.
And if you are willing to go to that effort, you will be rewarded for that in the contest. So, yeah, the spec will be posted tomorrow sometime.
There'll be some walkthrough videos, some more guides. If you have any questions, Ryan and I will be in the Q&A and chat for the rest of lecture.
And feel free to come to our office hours. Mine are Monday morning — Monday morning, and then Tuesday morning, I have some special project office hours.
Ryan can chime in with his if he wants. And, yeah, otherwise post on Piazza. Send us emails. Just please, ask a lot of questions, and we're here to help.
DAN GARCIA
Do you want to add anything, Ryan — Ryan T? RYAN T: Yeah, so just a little bit about lab 5.
Make sure to do that as early as possible because Logisim labs and projects take a little bit more time than coding things.
I have lots of lab office hours throughout the week on Tuesday to Thursday. And I think my TA office hour is on Wednesday as well.
DAN GARCIA
Perfect. Well, thanks to both of you. That's just wonderful.
BORA NIKOLIC
So one thing everybody should appreciate is the project.
And the project itself has been reshaped a little bit through the lens of folks who have gone through upper division classes recently that go through the describe process of design and really reflect that for masses now better — these are the guys that went through the ranks of 61C to 151
and 152, graduate classes, and understand the process, are very fresh to the process. And that is all reflected in the spec and in the design.
And when you come out of this, you have a processor in your hands. And don't do it just for the pair of AirPods. We'll try to get a few extra AirPods or whatever.
But do it so you'll have a RISC-V core in your hands.
DAN GARCIA
And as I said — BORA NIKOLIC: That's a huge thing.
DAN GARCIA
I don't want to sully the vote. But for many people, they find that the processor design is the most fun project in all of 61C.
We have some fun projects, and they find this is so cool to actually build this finally. So some people really get into it. That's wonderful. Thank you, Ryan.
That's delightful.
BORA NIKOLIC
Okay, we got to finish.
DAN GARCIA
Thanks so much, folks. Good stuff. You guys can stay, or you can relax. Thanks.
All right, let's keep on the topics left.
BORA NIKOLIC
We have eight minutes left to talk about it.
DAN GARCIA
I know we have a lot of stuff to do. All right, let me get in here. All right, here we go. So next, so this is the project 3 intro.
We did some of that. Bora and I took a look. We f to look at the responses.
There's a ton of them to read through, and we want to make sure we read everybody's comments. 761 of you replied, thank you so much for that.
That's a little bit more than half, maybe 60% of you, to respond.
I would say overall — I know, Bora, your first take on this. And my first reaction is these numbers were sobering. These numbers were not numbers that I was expecting.
What's harder is we don't have a baseline. We never did this during a non-COVID semester. So it's harder to know how people were doing. We used to do, hey, how was the midterm?
But it was less about your health, less about your mental health and workload.
So but we're hearing — you can see these numbers we're sharing with you now are just remarkable in terms of how overworked you are, how overtaxed you are, how mentally exhausted you are.
And we're not even on week 7 yet. We're just starting week 7.
So as we're just sharing all this, Bora, feel free to chime in with other thoughts you had about when we see these numbers. But I thought they were very, very sobering.
So how are you doing emotionally? We were hoping for 5's and 6's and 7's, and some of you were in that area. But most of you were averaging around 3.
So we had a little bit below 4 as the neutral. Most of you are feeling a little bit on the low side. And 7 is great, and 1 is awful.
Exhaustion was even stronger in terms of a signal for us to try to make some changes. Again, we're looking at 5's and 6's and 7's, and people feeling okay.
We do know, as we mentioned, that the early part of 61C is harder, and it does ease up at the end.
So we're now — the foot is — we're taking the foot off the pedal now with project 2 finishing up.
And then it's — only two projects still left for — I think there's eight weeks left. So it's much less of that, and it's thinking about letting it all sink in.
But the material comes fast and furious. And every three days is a new topic. That's positives and minuses. The minuses are it's a new idea to wrap your mind around.
The positives are if you got behind, you can rejoin and jump on the bus again and be part of it. So that's the part that's hard.
But nevertheless, this is, again, sobering in terms of how exhausted you are. And 7 is doing great, and 1 is just totally — batteries are dead at 1.
And we got many people at 1's, 2's, and 3's.
BORA NIKOLIC
Yeah. And my take on this is the first graph was not that bad. I was hoping for a 4 on the average. We got the 3.
DAN GARCIA
Yeah.
BORA NIKOLIC
But this one is — this is not a 100-meter dash, right? I mean, this course is not a 100-meter dash. This is more like a 5k race.
So we need to make sure that you guys can pace yourself. I mean, we have to not overwork you in the first mile. You have 2 more miles.
DAN GARCIA
And that was actually a comment or two by people in the overall general thing, which just, does the first beginning of 61C have to be this hard?
Does it have to kill us in the first couple of weeks?
Could it just — many courses, 162 and project courses, they get hard in the last five weeks, and the first five are setup. And then it's a big project that overwhelms you.
I mean, what's the — 150, Bora, right?
150 is this — you're in a class, where people are just drowning for the last eight weeks or seven weeks or something or six weeks for that project and living in the lab.
The little differences we're gonna see, we frontload a little bit of that, which it's lighter at the end. So it's a little different. And people said, does it have to be that way?
Could you just switch it back to there? We need to have a rethink.
None of you know this, but we had all the lead staff, head TAs and us, went through and did an estimate of how many hours per week with this.
And these were just very low. I think our estimates were much lower than they really are in terms of that. And also — BORA NIKOLIC: We made a few mistakes.
We made a few mistakes as we were walking. Yeah. And we are very conscious of that.
DAN GARCIA
And on top of that, this also reflects your exhaustion for other classes, too. This isn't just 61C, but 61C might be for many of you the thing that really is hard.
I do want to start the survey up, so let me get to the next one. Again, how are you juggling? That's very similar.
It's hard to manage — we got little bit less than less than average here. Prior knowledge, we were pleased.
We were thinking, well, maybe if it's the case that people came in without a strong 61A and B experience, and maybe then that's maybe the argument for maybe — oh, we need to hold the line stronger.
If half of you hadn't taken 61A or B or something and somehow snuck into the class, and the reason you're struggling is because of that. But it's not.
Most of you have completed 61A and B well and done well in that. And most of you — BORA NIKOLIC: And actually — DAN GARCIA: — C programming, either.
BORA NIKOLIC
Today, I did some — I tried to do a correlation, and there is clearly a correlation between how you feel about your 61A/B and how are you doing in this class or in general.
And there is correlation, but it's not that strong.
So there are people who did extremely well in A and B, but — or think they're very confident about their A and B, but they're still not doing great now.
So things happen. And we very much appreciate having — being able to know that.
DAN GARCIA
This is the time on the quest. We already knew this. We already knew this from the Piazza responses. But this is more of a — more data points here.
And the numbers we're seeing here — if you look at the average, it's around four to five hours for an exam that we were hoping to have be an hour, two max.
And people were mostly doing it in four or five hours, and a lot of people taking a lot longer than that. So this leads me to the next question I want to get to.
But let me — we have a quick poll. I want to get some sense above. But I want to just show you the rest of these numbers before we do this.
Project 1 timing — by the way, everybody did amazingly well on project 1A and 1B. The average — I think the median is 100 for both of those parts.
So people are doing very well, but they're putting the hours in as well. And I think we need to look at these numbers.
And then we had an estimate for how long project 1 should take. We need to look to see if it's at all online. But a lot — there's some high numbers there.
BORA NIKOLIC
So, yeah, and the estimate was around the mean. But the problem is we have a very long — DAN GARCIA: It's a long tail. It's a long tail.
Yeah.
BORA NIKOLIC
Yeah, a long tail, even if you throw out some of those that may not be completely honest answers. This is a very long tail.
And this did hit some people, and not necessarily because of 61A or B. It was just things that had to do with providing tests and things like those.
DAN GARCIA
And part of our perspective is we thought that people learned how to do tests in 61B. That was the point of that.
But we were hearing loud and clear that people are really not ready to do tests yet.
There might be a breakdown in A and B of preparing students for C, where we're thinking that they're ready to do their own tests, and in fact, they're not, possibly.
So that also can reduce the workload if we give you to set more tests. Overall, you're supposed to be spending six hours per week.
It's 12 — 12 hours per week for a 4-unit class, and the numbers are much higher than that that we're seeing.
But there are a fair number of folks who are in the lower numbers, but a big chunk of 15 and a big chunk of 20 and an even bigger — blips of 20 and 28.
And that's inappropriate in terms of our point of view. We need to fix this. We need to be better. And we need to not be overloading as well.
So again, this is all things for us to fix. How many of these we can fix — they may have to fix in the spring.
But we're going to see what we can do to try to compensate for this here. So to this end, any midterm coding problems?
There's going to be a chance to do some coding in C, a smaller problem, to replace your quest. We have to do that. And there's some coding in MIPS.
We are going to have all test cases given to you. We're going to reserve some of those to prevent you from hard coding.
If we give you every single test, you could just basically look at our tests and just get it working for those. We're going to try to support you there.
But we're going to have a couple — example, split. You were supposed to do this splitting thing for your thing.
If we have a similar thing, we're going to obviously have a big, long one and some points for handling our hidden — one hidden test. But mostly, we're giving the test to you.
So that's very fair. But we have to obviously have a couple of them, so you don't just hard code to that. We will announce how long it took all our TAs.
We will survey our TAs and announce that in advance, as well as announce every piece that's on every exam. You're going to have six questions. They'll be on this topic.
You already know what the topics are. They're everything up to — up and through pipelining. But we'll make sure that you know. And you are going to be coding, so be ready for that.
But we will give you your tests or a fair number of those tests. We're going to allow you for 48 calendar hours to take the exam. That's another piece of it.
People said, we don't — I was busy on Sunday. Even though we announced it, well, that was a bad day for me, either because of religious reasons or other scheduling things.
Give me more flexible time. So we're going to have one weekend day and one weekday.
And I have a poll right now that we're launching that I'd like you all to fill in two questions we're asking you. Number one, what days would you prefer to take it?
We're going to probably give a 48-hour time limit for what we believe is a two-hour exam that may be there. So that's what we're aiming for. The poll is open.
Please, do click on the poll if you can. I see nobody's voted yet.
Please, jump in on the poll, so we can see those numbers, and we can see when people are taking it, because we have to staff Piazza to answer these questions.
The second question is we are of two minds of whether we give you all 48 hours and say it really shouldn't take you more than three — two, three, or four hours max, we believe.
But do we give you all of this? Some people said, if you give me 48 hours, I will spend 48 hours. You have to stop me from my best interest.
So help me limit that. Counter to that is some people take longer. When I was in college, I took longer to do exams. I did well.
I got to Berkeley. But I took longer to do exams than my neighbors. I would just be more refined and quieter, and I would just make sure I was more perfect with my answers.
So I want to react to that. For the students who need more time — I think it's more fair to give them the more time.
But some people are saying, no, don't have an unlimited time thing. So that's at issue. So we wanted to get a sense of that. And keep on voting.
We have about 92% of you voting. And what's funny, Bora — you can't see these. Only I can see these. These are very funny.
BORA NIKOLIC
I see them. I see them.
DAN GARCIA
Oh, you can see too? Look at the second one. Let me share this with you. This says, no signal, whatsoever.
Let me close this — 96% of our ending poll. I'm going to share the results with you. And what we're seeing is you're all across the board Friday, Saturday, Sunday, and Monday.
Most of you are — Saturday and Sunday will work for when you're going to do it. But some of you need a Friday or Saturday. But we need to limit.
We can't just have it open for four days. Our staff needs to be able to do other things.
If you have the four days, maybe it'll be a longer time for you to get an answer. So that's harder if we're not staffing it all the time for that.
The second one I see here are just evenly split whether we should not have an upper limit.
I'm seeing 45% say no upper limit, and a little bit less than 40%, maybe exactly 40-ish, 41%, on, yes, upper limit.
I don't know what happened.
BORA NIKOLIC
Similar situation here.
DAN GARCIA
What happened to your data here?
BORA NIKOLIC
It's the same situation about these little clicker questions. We are going to make exactly half of the class unhappy.
DAN GARCIA
Unhappy, exactly. Exactly. And whatever way we make unhappy, I mean, if we had to lean in one way, I would lean toward no upper limit, just because you can satisfy more.
But we would ask those who need an upper limit to set one internally. That's what we'd ask. If we end up leaning that way, we would ask you, please, just set a timer.
And stop working on it after two hours or an hour or whatever. We will probably break this up into pieces like we did last time.
And we're going to do the same proctoring as last time. And hopefully, people have practiced their proctoring stuff so that they — and it'll be smaller times. It'll be pieces.
It won't be a long two-hour video. We have shorter pieces. So I think we'll have probably — I think we said, Bora, five questions or so and five small pieces for that.
And you'll stop and start your video for that.
BORA NIKOLIC
Yeah.
DAN GARCIA
All right, I'm all done.
BORA NIKOLIC
Okay, so we are well over our time.
DAN GARCIA
Yeah, yeah, yeah. Thank you so much, folks, for that feedback. This was a fascinating poll with no data. But again, good luck with this.
Continue with this. We're at week 7, and we'll be back next week at week 8. Any last thoughts, Bora?
BORA NIKOLIC
Labs and then projects on — from tomorrow. All right.
DAN GARCIA
Good stuff. All right, folks, take care. Thanks so much for attending our live session, folks. Wonderful.
Bye-bye.
BORA NIKOLIC
Thanks. Bye.
Đăng nhập để đánh giá học liệu này.
Đăng nhập để bình luận — chưa ai bình luận về học liệu này.
Không có khóa học liên quan.