Risking RISC-V, Engineers & the gig economy, chiplets
I’m Brian Santo, EE Times Editor in Chief, and you're listening to EETimes on Air. This is your Briefing for the week ending June 14th.
我是EE Times主编Brian Santo，你正在收听的是EETimes全球联播。这是截至6月14日的一周热点新闻概要。
Today we’ll be talking about…
RISC-V, a free and open processor architecture. We examine the potentially profound consequences RISC-V might have for the smartphone market…
We’ll also be talking about engineering education. We’ve heard about the need for STEM education, but the required skillset for engineering has expanded in unexpected ways.
We’ve also got a report on chiplets. With Moore’s Law coming to an end, one way to compensate might be to create a system whereby manufacturers can assemble a microprocessor from a set of processor subsystems-- chiplets.
Articles Discussed in This Episode:
Qualcomm Takes Stake in SiFive 高通投资SiFive
Engineering Jobs & Engineering Education 工程类职位与工程教育
Chiplet Effort Plays First Proposals Chiplet开放标准的第一次提案出台
Andes Technology Launches FreeStart to Boost RISC-V Adoption 晶心科技推出FreeStart项目以提升RISC-V普及率
First up, RISC-V. Qualcomm is a leading supplier of the chips that go into smartphones. European correspondent Nitin Dahad wrote a story about Qualcomm investing in SiFive, a hard-charging little start-up that has expertise in RISC-V technology. International editor Junko Yoshida caught up with Nitin to discuss what RISC-V is and what Qualcomm’s investment in SiFive might portend.
首先来谈谈RISC-V。高通公司是智能手机芯片的领先供应商。EETimes欧洲通讯记者Nitin Dahad撰写了一篇关于高通投资RISC-V技术初创公司SiFive的文章。我们的国际主编Junko Yoshida与Nitin一起讨论RISC-V究竟是什么，以及高通对SiFive的投资可能预示着什么。
Hi Nitin! How are you?
I'm fine, thank you. I've just arrived in Zurich.
What’s happening in Zurich?
So the RISC-V workshop is happening here in Zurich. So I've decided for the first time to attend one of these. So I'm going to be attending it over the next couple of days.
Excellent. All right, so I think I've come to the right person to ask this question, because I think last Friday you filed a story about SiFive getting Qualcomm's investment. So Qualcomm’s investment in SiFive, what does it tell us about the state of RISC-V, and more importantly, about SiFive?
I think it means RISC-V is coming of age. I had a conversation at Computex in which a major chip industry investor was asking about the merits of RISC-V and whether it really can be deployed more widely. What I think this investment from Qualcomm now does is, it's publicly saying-- well, actually, Qualcomm did publicly say it wants to exploit the full potential of the RISC-V architecture in wireless and mobile. And I think a key thing for the RISC-V architecture.
Right, okay. So this is my opinion, but it seems like Naveed Sherwani, president and CEO of SiFive, he is a tireless promoter of his own business. Nothing wrong with it. All the more power to him. But in your story, you quoted him claiming SiFive having just achieved 101 design wins. I read in LinkedIn today he was talking about the 102nd design win he is onto or something like that. Is SiFive a chip company or an IP company? I'm a little confused.
Just from what I remember from my days at ARC, I think you can sort of put the numbers out. A design win can mean it's very advanced, it might be very early stage or it might be just some concept.
So anyway, coming back to your question. This is where the perception seems to flip-flop. Some people tell me SiFive is pushing IP, and Naveed has been talking about both.
But when I most recently spoke to him in Cambridge in the UK, he categorically told me he wants to be a silicon company. He told me they’ll do $100m US dollars in revenue this year, and 75% of that will come from silicon and 25% from IP licensing.
Wow. I did not think that. That's interesting. So I know that Naveed doesn’t mince his words when he talks about current IP licensing models out there on the market today. What problems is he seeing in it, and what promises is he making to fix that problem?
His core message is you can do a chip very quickly. He says the reason you can't at the moment is it takes too long to negotiate IP license contracts, say with multiple vendors if you've got a chip with multiple IP blocks. Also if a designer needs to make changes, then it can often require re-negotiation of the license or licenses.
He claims SiFive’s model addresses this head on, with all the IP its IP partners residing within the templates they provide in the Cloud. So the designer doesn’t need to negotiate licenses in the first place with multiple IP vendors. So that sort of accelerates his ability to do the chip.
I see. I'm just wondering: Am I the only one thinking that RISC-V Consortium is kind of overshadowed by SiFive, or is SiFive stealing all the thunder from RISC-V? What do you think? What's your take?
It's very interesting, and I'm going to take it to another level. SiFive is certainly stealing all the limelight, spending a lot of its $125 million US or so that investors have put into it on its traveling roadshow in almost every corner of the world (over 50 plus events).
I am sure the investors are happy with this profile building. This in turn raises its valuation for an exit. That’s the norm with companies that raise a lot of money. Look at Graphcore in the UK. Companies like this, they raise a lot of money looking for an accelerated path to exit.
But in RISC-V, I think there are many other emerging vendors, and I think we’ll see them also raising money and profile on the back of this SiFive funding.
I hope to meet a few at the RISC-V workshop here in Zurich, Switzerland this week.
All right. So we look forward to your follow-up stories. Well thanks for your time, Nitin.
After recording the segment you just heard, Nitin wrote another story on a company called Andes Technology, one of the RISC-V specialists offering engineers a way to try RISC-V development for free. Read it on eetimes.com.
The world needs more engineers. We expected that would include more IC designers, but we were surprised. This week we presented a set of articles that examine Engineering Employment and Engineering Education. Our main article finds out that in some categories, the electronics industry needs fewer design engineers, not more.
What does that mean for the engineering profession moving forward?
We’ve been told that we need a laser focus on STEM skills: science, technology, engineering and math. But as we investigated, we discovered the electronics industry needs a much wider variety of skills, including communications skills, collaboration skills, business skills, and subject matter expertise. And all of that is beginning to get integrated into engineering programs.
George Leopold is our editor based in Washington DC. His contribution to our Employment and Education Special Project was on that very subject, so we invited him on for a chat.
So how you doing, George?
Good, Brian. Good to be with you.
You recently took a trip out to Olin College. Tell us what you found there.
Well the school was started about 1997, and the whole idea was that the people that founded it thought engineering education is probably too narrow. We need to expand it. And not only did they expand beyond math and science, but they added things like arts and humanities. So that's certainly a different twist. But I think they see that as a way to promote more innovation, particularly as a lot of their graduates move into software areas.
Small school. Only 90 students per class, 360 total. and they've got a heavy emphasis on ethics and context as well. They wants their students to think about the consequences of the technologies that they're creating. Unintended and intended consequences.
So we think it offers a nice framework for possibly the future of education and engineering.
Well that fits into some of the coverage we’ve had over the years here at EETimes over the past few months, talking bias. Bias that might be built into the product itself, but especially with artificial intelligence, where the assumptions that you bring for training the AI might have some inherent bias.
So I imagine a program like what they have at Olin would help focus people on that set of concerns.
Yeah. That's a primary example of I think the types of things they want to address in their training. One of the professors said their introductory math courses are almost like an architect's studio. So they take a very different approach. And this stuff they emphasize is baked into the curriculum. It's not a bolt-on. And that's been a problem. Some of the education studies that have come out over the last year, that it's kind of thrown in. But these guys actually want to make it integral to their curriculum.
Again, I think it's something that's probably generational here. We're moving away from sort of the standard engineering curricula to actually looking to see how this stuff's going to be used and what are the intended consequences, AI being a classic example.
So there’s an expectation that a lot of what we’re going to be needing moving forward in AI, machine learning, it's going to be a lot of coding, a lot of software engineering. The jobs look like they might be going to software. And it's interesting what you just said about the curriculum at Olin. It sounds like with liberal arts integrated into it, when you're doing AI, you’re going to need subject matter experts, right?
Yes. They want to broaden people’s perspectives. We have a contributed article by an assistant professor at Olin, a woman named Alison Woods, and she told me her background is in theater arts. And now she's teaching engineering. So that sort of gives you an idea of where they're coming from. Very different from sort of the MITs and Stanfords of the world.
Another thing I wanted to point out, Brian, is that we seem to be sort of missing this Sputnik moment in American education. In '57 the Russians launched Sputnik and we realized we've got to catch up. So we really emphasized math and science, engineering, STEM. In some ways we've found in some of the stories that we're doing for this package, that STEM has just sort of become a slogan. And I don't know that we've necessarily delivered on it. Maybe the E in STEM should stand for ethics as well as engineering.
It also sounds like if the Olin approach seems to be the way engineering is going, an emphasis on STEM might be inadequate.
Maybe perhaps it should be as Dave Vasco from Rockwell recently told me: Maybe it should be STEAM. Add an A for the arts.
So let's look at... Let me ask you about this. What should we look at with engineering moving forward? Specifically engineering education. Should it focus on software, AI, ethics, liberal arts, collaborative abilities, communications, the ability to work in teams and be able to tell people what it is you’re doing with what you’re making?
Yeah. A heavy emphasis at schools like Olin. In fact, they look for collaborative skills, the ability to work on teams, to work on projects like robotics or systems integration, solving the right problems. Not just... Identifying the problem and coming up with a solution. Those are the types of things they're trying to bake into their curriculum. And I think we'll see a lot more of that.
The other thing of course is, it's like technology, this approach has to be scaled up. Right now it's only being used at smaller schools. Olin only offers undergraduate degrees, so the trick will be how you scale this up to big schools-- MIT, Stanford, the big engineering schools, their graduate programs, their post-graduate programs as well. Although I'm told that a lot more engineering students aren't going to grad school because the job market's pretty good. And I guess one of the questions we raise is, if you do get a job, are you going to get a full-time job or is it going to be a gig? That's another issue for us.
Absolutely. A lot of different changes in engineering, engineering employment and engineering education all coming up.
Thank you, George, for being here with us.
Good to be with you, Brian.
And we'll have you back soon.
Moore’s Law is petering out. One of the more curious approaches to keeping Moore’s Law going is called chiplets.
It used to be that if a system manufacturer needed a processor, its best option was to buy a big, highly integrated monster of a chip from one vendor. But now manufacturers all have specialized needs. One size processor does not fit all.
The idea behind chiplets is for all the chip vendors to break down their big monsters and make processor subsystems instead. Those are the chiplets. That would give the manufacturer the option to assemble a processor out of the chiplets from different vendors as best suit its needs. Each assemblage of chiplets would act as if it were one single chip. That’s the idea anyway.
How far along the chiplet path are we really? Dylan McGrath has the story.
This week I had the chance to drop in to attend a workshop of the Open Domain Specific Architectures industry working group, which is part of the Facebook-backed Open Compute Project. This group is working on advancing an open chiplet ecosystem and has proposed a protocol for chiplet interfaces.
Not having followed the day-to-day development of this industry group or the growing pains of the open chiplet ecosystem movement, I was on one hand surprised by how far along this development seems to be in many ways, but on the other hand I was really kind of awestruck by how far this still has to go before we get anywhere near the sort of utopia of a future where people are just kind of taking different chiplets from different sources, manufactured by different foundries, in many cases at different technologies nodes, and kind of slapping them together in a plug-and-play type of way into a single package.
There really is so much that is still needed, including standards, tools and really entirely new business models to make this all work.
There are a lot of questions about how companies involved will be compensated, who will be responsible for what. There's a long way to go.
But I had the opportunity to sit down at the workshop with Ramune Nagisetty, who is a senior engineer and director of process and product integration for Intel's technology development group. She has really led much of the work being done at Intel around chiplets. She enlightened me quite a bit in many ways, including describing what she believes are the two most significant technical hurdles to an open chiplet ecosystem: testing and the ability to ensure known good die, and the thermal challenges associated with cramming more board-level components into a system-in-package.
Despite all the work that clearly needs to be done, I was very surprised to hear Ramune say that the whole development of an open chiplet ecosystem is really coming together very quickly and is going to look a lot different over the next three to five years. She told me that the ramp is very slow, but it's going to increase in pace as each standard and tool is put into place.
Ramune and a number of other prominent technologists that I met at the workshop are very bullish on this chiplet ecosystem. That is something that will be in place over the next handful of years. And of course this could dramatically change the way that advanced devices are created.
Dylan also collaborated on a story with Barb Jorgenson, our colleague from EPS News, about the trade war with China.
Huawei is one of the biggest customers for a lot of US high-tech companies. Dylan and Barb report that financial analysts are saying that the Trump administration's ban on Huawei is going to hit the Chinese company’s many US suppliers where it hurts the most: their bottom lines. Investors are spooked, too; the stocks of some of those US companies are getting dinged.
Since Dylan and Barb filed their story, Reuters reported that Tokyo Electron, a major global supplier of equipment used to make chips, will comply with the ban. President Donald Trump threatened Chinese President Xi Jinping with more tariffs if Xi declined to meet with him at the G20 meeting coming up in Japan, even though there was no indication from China that Xi wouldn’t.
Secretary of State Mike Pompeo threatened German Chancellor Angela Merkel that the US would cease sharing intelligence with Germany if German companies were to do business with Huawei.
Huawei announced it would press its patent rights with Verizon, saying the US communications giant should pay it a billion dollars for use of over 230 Huawei patents. Ladies and gentlemen, this is going to get worse before it gets better... if it gets better.
This week’s bit of tech history:
Forty years ago this week, Texas Instruments introduced its Speak & Spell talking educational toy. Last week I just happened to be chatting with one of the Speak & Spell’s inventors.
Gene Frantz is now leads a company called Octavo. He told me he and his three coinventors just got lucky getting the development of the Speak & Spell funded. While they were building it, they discovered literally nobody wanted it-- not TI managers, not educators, not parents. Nonetheless, they got to finish the project because TI management neglected to create a clear reporting structure for them, so they were all but freelancing. And when they did get hauled in to report on their progress, they were the only team in the company coming in under budget. Kids, it turned out, loved the thing.
The Speak & Spell is indisputably one of the greatest products in electronics history, helping to propel TI to an early leadership position in the new discipline of digital signal processing.
And here’s a bit of audio from the Speak & Spell.
SPEAK & SPELL:
Now spell EETimes On Air. That is correct. You win.
All these year’s later, it turns out it’s got your number.
And that was your Weekly Briefing for the week ending June 14th.
This podcast is Produced by AspenCore Studio. It was Engineered by Taylor Marvin and Greg McRae at Coupe Studios. The Segment Producer was Kaitie Huss.
I’m Brian Santo. Catch us here next week at EE Times On Air.