DAC News, Wednesday

The last day of the DAC tradeshow is the best...said nobody ever. After two days in the booth for exhibitors, or two days watching non-stop demos and presentations for attendees, the third day seems endless. Although the exhibition finishes on Wednesday, DAC itself continues with management and technical training on Thursday, following an opening keynote about autonomous robots. For programmers, there is advanced C++ training. There is a management training track, SystemVerilog, and more.

If you are at the show and have visited the Cadence booth, then you already know we have an Angry Bird theme going (new bird every hour and a half, your kids won't forgive you if you don't get all eight). As is now traditional, that theme continued at the Denali party, where CEO Lip-Bu Tan and VP Marketing Nimish Modi danced with Disco Inferno dressed in Angry Bird costumes, along with a dozen local kids dressed as the green pigs.

Keynote: The Challenge to Create Truly Great Products

Wednesday's keynote was by Mark Papermaster, the CTO of AMD. His background includes time at IBM and Cisco, and at Apple, where he worked on both the iPod and iPhone. Given that track record, I think we can say that he knows something about how to create "truly great products."

Mark talked about the three eras. First, PC and the internet. Innovation was largely driven by Moore's Law. As he says "I just took it for granted that every couple of years we could have more performance for the same power and price." Next was the era of the smartphone and apps, which has had meteoric growth the likes of which have never been seen before. But Moore's Law is slowing down and there is no entitlement to do innovative things and know that at same cost and power we can move our ideas forward.

We need to change the way that we think about design. If you need full mobility, you will continue to see monolithic integration on big SoCs, but it needs very high volume. For applications that need high performance there needs to be lots of innovation as to how to keep performance improving on Moore's Law pace. This will create the next phase, the immersive era, involving CPUs, GPUs, and accelerators.

A number of trends are coming together to drive this. Pervasive connectivity at higher and higher bandwidth. Machine learning and recognition has advanced at incredible rate. Voice recognition accuracy is also improving fast. Graphics rendering on ultra-high resolution displays. And, of course, software. Especially open-source software.

Every year, Mark goes to CES and sees lots and lots of innovative products. But he knows from history that most of them will be commercial failures.

So what makes a great product:

• Meaningful problem solved
• Impactful innovation
• A better experience

Some examples:

• iPod and iPhone: How do we create a different experience?
• Game consoles: User experience matters
• VR head-mounted displays: Predicted by 2020 over 200M will be sold

However, to make it happen, you will need to bring a team together. The "what" and "why" is not enough, and many products fail in the "how" stage. Or as Antoine Saint-Exupéry said rather more poetically, "A goal without a plan is just a wish." So you need:

• Clear goals and plan
• Right team and culture
• Maniacal execution

Many years ago at a DAC keynote, Cadence CEO emeritus Joe Costello gave one of his rules, "Write the press release first." Mark had almost the same rule as a way of keeping scope-creep under control and keeping the design simple, "Write the press release now."

What is it going to take to make the immersive era a reality? This should be the year it all starts, although it will be a slow ramp. But it could be big. Bloomberg reckons it will be $80B by 2025. It starts with entertainment and gaming as early adopters. Then training and simulation (Royal London Hospital recorded an entire surgery and allowed training students in a VR environment). Medicine. Education. Big data visualization. Remote presence.

All of this requires big increases in both CPU and graphics performance. Of course, that is just where AMD plays. Their technology is being driven by the drive to artificial intelligence and machine learning, and by the drive to photorealism. Today we are 90 FPS at 2160x1200 (2K) with <10ms latency. In the future, this will go to 144 FPS, 15360x8640 (16K) with <3ms latency.

Final message: “There’s never been a better time to develop great products."

Fireside Chat with Cadence's CEO—With Real Fire

Appearing without his Angry Bird costume, today it was Lip-Bu Tan's day for an unscripted interview with Ed Sperling. Usually the phrase "fireside chat" is just a figure of speech, but today's came with a real fire. A spotlight caught fire above Mentor's booth, and the entire conference center was evacuated for half an hour or so. Luckily, we got the all-clear and Lip-Bu's chat started only 15 minutes behind schedule.

This blog post has got so long that the details of the chat will be in tomorrow morning's Breakfast Bytes.

Security Netflix-Style

Bryan Payne leads the platform security team at Netflix. (They are headquartered in Los Gatos in what used to be the garden center where I would buy plants and stuff when I lived in Los Gatos myself. Well, they closed the garden center, built a big new building, and suddenly it was Netflix.)

Bryan started by telling the story of the Antwerp diamond heist as an example of overcoming multiple layers of security. It is a fun story but too long to repeat in full here. Plus, there is no need, Wired Magazine already did a great piece on it.

Physical security varies depending on what you are trying to protect. An area of Antwerp through which billions of dollars of diamonds pass, a lot. Bank vault, a little less. The lock box at home, a lot less. It is the same in the computer world. NORAD doing missile command and control, ultra secure. Google or Facebook datacenter, maybe not quite as much. Home laptop, maybe not secure at all.

A typical computer system these days has hardware at the bottom, probably a firmware level, an operating system like Windows or Linux, and then application software like a web browser or web server. But there are vulnerabilities in places you might not think of. There has been evidence of attack by inserting Trojans into the firmware of hard disk controllers, disclosed in some of the Snowden documents. This is very hard to do and very hard to detect. It's not as if you can run a virus check on your hard disk drive firmware. But it has access to all the data on the disk, so has the potential to count as a major breach.

The next level up is the OS. A rootkit is software that operates at this level. It does two things. Firstly, whatever it wants to do. But it also adjusts the operating system and its data to hide its existence. If you list all the processes running on the operating system, the rootkit process will not appear.

The application level exists in two forms, on the clients (such as your computer or your smartphone) and the server (such as Google or Netflix).

Security is a bit like terrorism: the good guys have to get everything right at every level all of the time. The bad guys only need to find one weakness somewhere. The attackers can seek out the weakest link. That link might be somewhere obscure. The code that was inserted into Juniper's source code altered the random number generator in a way that made it more predictable so that the cryptography could be attacked directly.

But the real world is complex. You typically don't interact with a single server, it just looks like it. Netflix, for example, has thousands of little services operating in the cloud (they don't operate their own datacenters, they run on Amazon). In between are routers and switches too. One reason security people love encryption is that you can encrypt end to end and not need to worry about anything in the middle since the encrypted data can't be compromised.

The reality is that we live in a world with millions of things, and solutions need to scale. Bryan had another real-world example to make it clear. If he gave each of us a penny and asked us to determine if it was fair we could probably do it. Toss it 100 times, maybe even build a machine to toss it. But security is more like being given a million coins and asked to find the one that is not fair. It is hard to scale solutions to this level.

One other reality in security, the attacks never get easier. Every year new approaches are invented, new vulnerabilities are found. You can't uninvent them.

The three biggest problems Bryan sees in the space are:

1. Everyone is deploying software in the cloud (Netflix, too). They assume that the cloud provider has done security right but there is no way to verify that. Two useful changes for the future would be either to be able to verify cloud security, or else develop approaches that don't depend on the cloud not being compromised.
2. Complexity is the enemy of security. It needs to be easy for people who don't have a PhD in computer security to get security right through simple-to-use libraries and so on.
3. There is not enough talent. Not enough people are trained and talented. Linkedin did an analysis and found there are three or four job postings for security engineers for every actual security engineer. In other words, the demand for security is three or four times the total amount available. No wonder systems get compromised.

In the questions at the end, Bryan was asked what they did about security when they moved to Amazon. Netflix has a dedicated machine that they are the only users of. That is where they can put crypto keys and have reasonable confidence that nobody else can get there even if Amazon itself is compromised. This seems like the cloud version of what we do on chips, where we put the keys in special hardware areas and don't rely on the software on the as the only protection.

Art Show: Grand Prize

This year for there was an art show/competition. There were various prizes but the grand prize went to Coventor for their model of 14nm FinFET transistors. David Fried, the CTO of Coventor explained how they did it. First, they used their own product, Semulator-3D to generate the data. Normally this is just rendered on the screen. But they used a state-of-the-art 3D printer to "print" in three colors with a support material. Since they only had three colors, they had to work out how to dither to make all 27.

There is a Cadence connection, too. Back in the mid-2000s, Cadence had a venture fund called Telos that invested in Coventor. Jim Hogan and Charlie Huang were among the general partners. They also invested in Virtutech where I worked at the time. I believe Cadence still owns a small part of Coventor. So we can bask in reflected glory at our partial art prize.

Next: Lip-Bu's Fireside Chat with Ed Sperling—With Real Fire

Previous: DAC News, Tuesday