Maybe it's time for Broadcom Chief Technology Officer Henry Samueli to get a little of the limelight.
For years, CPU makers like Intel, IBM, and Apple got the glory in the processor industry. But as mobile and cloud computing explodes, communication chips are getting more important, and that's the market where the company Samueli co-founded is a dominant player.
Wednesday, the company announced a new processor that's emblematic of the new trends in the computing industry. The BCM20736 is a modest little thing -- a chip that lets a device communicate over low-power Bluetooth links and charge its batteries wirelessly. But the chip fits neatly into two hot new trends: wearable computing and the Internet of things.
This first category includes fitness devices, Google Glass, and assorted smartwatch efforts. The second involves spreading the Net beyond PCs and phones to things like, , , , and, yes, .
Samueli founded Broadcom in 1991, a time when wireless communications were a rarity in the industry and unheard of for consumers. Now electromagnetic waves carry not just data, but power, too. But wired communications remain critical to the overall Internet.
Samueli shared his vision for communications technology in an interview with CNET's Stephen Shankland. Here's an edited transcript of a conversation that touched on wireless charging, LTE mobile networks, gigabit broadband for the home, and more.
Q: You have a new Bluetooth chip with wireless charging. Where do you think wearable computing is headed, and where do you think the Internet of things is headed? It's hard to sift the hype from the reality.
Henry Samueli: There's enormous hype, but for good reason. It's actually a pretty exciting market because it really is hundreds of markets in one. When you talk about the Internet of things, you're covering hundreds of industries. At the core of it is communications. That's the piece we supply -- this ultra low-power, low-cost communications chip that you put into whatever flavor of thing that you happen to want to build. The nice thing about this Internet of things market is that the barriers to entry are fairly low. You don't have to be a monstrous company. There are going to be thousands of startups created to deploy all these devices.
You buy a simple solution, like from Broadcom we have this whole product line called WICED -- wireless Internet connectivity for embedded devices -- it's a very easy-to-use platform with software that you can connect up to whatever type of sensor you may find out there. Whether it's an environmental sensor, a medical sensor that measures bodily functions, a sensor in your home for security, you need to connect it to a communications device. Then it goes out to the Internet.
It costs a lot to lay a fabric of sensors all over creation. It's a big investment if you want to wire up all your roads and traffic lights and streetlights and houses and power meters. Where are the places where this is going to catch on? What are the markets where it's financially justifiable at the outset?
Samueli: Infrastructure is going to take time, but it's going to take time. Every meter will be a smart meter -- you'll have sensors everywhere in the environment. The shorter-term opportunities are consumer gadgets that people will buy on Amazon. We're starting to see a lot of it in the health sector. It's more of a fitness craze where you're wearing wristbands that monitor your steps. Eventually you'll have more sophisticated sensors that are monitoring your heart rate or other bodily functions.
Smartwatches is another interesting opportunity. We're seeing a lot of major companies as well as startups coming up with smartwatches that replicate a lot of the functionality you might have in your smartphone. Will it be as big a market as smartphones? Probably not, but it still can be a very substantial market.
And then, in the home, we're starting to see a lot more wirelessly connected appliances. Smart thermostats, smart smoke detectors, and all the flat-panel screens everywhere talking to all your tablets and smartphones and providing full multimedia distribution around the house.
What about getting power to all these devices? You're with the Alliance for Wireless Power (A4WP). Can you use that to get power to anything and everything? Today it's a very short-range technology: You have to set your device down on a charging pad.
Samueli: Currently, yes. But there are people looking at longer-range wireless power. You can direct the wireless power toward a device over a much longer distance. That's still very experimental, but over the next few years, you might have the ability to do wireless power at a larger distance. I think that's ultimately the way you're going to charge many of these devices. Or if you can get the power dissipation down far enough that you can run it on a coin cell [battery] for a year, you might be OK with that.
You've signed up for A4WP. Wouldn't it be nicer if we just had one wireless charging standard?
Samueli: We have to get to one, and they're talking. There are three consortiums out there. They are now actively speaking to one another to come up with some convergence. You can't create a mass market if you don't have a common standard. If you lay your device down on a pad and it doesn't charge, you're going to be very upset. It has to converge, and I think it will.
Why did you go with A4WP?
Samueli: We're very customer focused, and customers were pushing us in that direction. We also could very easily do the other approaches as well. They're all fairly similar in terms of the core technology, and you can even do a multimode chip that can accommodate all three.
Then there are energy-harvesting ideas, using the kinetic or thermal or solar or radio energy in a device's environment. Is that good for anything useful, or is the amount of current too miniscule?
Samueli: It would have to be a very low-level sensor to be able to survive on harvested power. I'm sure there will be some applications for it, but I don't think that'll be a mainstream market segment.
I'm curious about your thoughts about LTE wireless networking for mobile phones. You guys bought the LTE assets from Renesas Electronics. You've been a laggard in the LTE market, so will this catch you up?
Samueli: The deal closed October 1. The intent there was to accelerate our time to market for LTE. The Renesas team, which is originally the Nokia modem team, has been working on LTE from the very beginning of the early standards days in 2002, 2003, and 2004. They've been developing the technology for over a decade, so it's very mature. It's been carrier-qualified and carrier-certified. We're very excited about finally entering the market for LTE.
Can anybody catch Qualcomm there? What's your go-to-market strategy going to be -- now you have a second supplier? Or is it going to be more aggressive?
Samueli: It's important that the market has multiple suppliers for this kind of technology. Anytime you have a market this large, the customers demand multiple sources for their technology. The opportunity is huge to enter this market. It's a matter of having a complete set of solutions, not just the LTE. Broadcom has a leading position in connectivity -- all the Wi-Fi, Bluetooth, GPS, NFC (near-field communications). Combining that with the LTE baseband allows you to develop a full platform solution for your customer, which gives you a very powerful position. We do believe we have the ability to make a very significant play into that market.
Why did it take you that acquisition to get you there? LTE has been around for quite a while, and Broadcom is right at the heart of wireless communications. Why didn't you have this in-house?
Samueli: LTE has accelerated faster than most people had anticipated. It really took off very quickly from the time it was introduced. We did have our internal development road map, but we just needed to accelerate it.
With LTE, as I understand it, if you want to get the real high data rates, you have to use higher frequency bands, and in order to do that, you have put your base stations closer together because high-frequency signals attenuate faster and the transmission range is shorter. To meet the potential for LTE, will wireless network operators have to build a bajillion new towers?
Samueli: We're projecting forward now 10-plus years. For the next decade, the current infrastructure will serve us just fine because LTE data rates get you to several hundred megabits per second with LTE and LTE Advanced. But if you look out 10 years and beyond, where you're going to need to get multigigabits per second, then yes, smaller cells are going to be required, so there is going to have to be fairly major infrastructure expansion.
You do network equipment for fixed-line communications, too, not just wireless. What do you think about fiber-optic links to the home? You see Google trying to push Google Fiber. Is that ever going to be big or cost-effective?
Samueli: It will. The drivers are services demanded by the consumer within the home. You have to look at the Ultra HD [high-resolution 4K video] that's just been deployed. You're going to have higher data rates for digital television and high-speed Internet access that consumers want. You're going to need gigabit data modems -- cable modems, DSL modems. Fiber-optic transmission will be a part of that with passive optical networking. Broadcom is a market-share leader in every one of those markets. Driving higher data rates to the home because of consumer demand will absolutely happen.
What's your forecast for 10 years from now? What fraction of homes in the US are going to be hooked up with fiber? I imagine it'll still be pretty low. I'm amazed how much people keep milking out of copper wiring.
Samueli: The current infrastructure is primarily dominated by the hybrid fiber-coax networks. The cable companies deliver fiber to a node, and then the node splits off over coaxial cable to your house. The consumer just sees coaxial cable. That hybrid structure will last for a long time -- certainly 10 years if not longer. The penetration of fiber will still be low for a while.
How fast can you get the coax cable infrastructure to go?
Samueli: Right now we're working on the next generation of the standard called DOCSIS -- the Data Over Cable Service Interface Specification. DOCSIS 3.1 is being defined. That will allow somewhere between 5 and 10 gigabits per second delivered to the home. That is a very high data rate that will satisfy the needs for the consumer for the next decade easily.
What are the constraints on that working?
Samueli: Very little. We're designing it to be compatible with the existing coax cable infrastructure that's out there. All you have to do is upgrade the cable head end [the cable company's network gear] with a new box that supports DOCSIS 3.1, then you'll have a DOCSIS 3.1 cable modem and you'll have 5 gigabits of data rate capability. It's going to happen soon -- probably within three years you're going to start seeing deployment of that in a fairly significant way.
Is the 802.11ad, 60GHz Wi-Fi technology going to be good for anything, or does it not have much utility since the super-high frequency means the signal range is so short?
Samueli: It's a trade-off -- range vs. data rate. If you really want 5Gbps to download a file extraordinarily quickly and only need it to run over 1 meter, then maybe you'll use the 60GHz technology, so I wouldn't write it off. There may be interesting applications for very high-data-rate, short-range wireless, and I think 60GHz has potential over the next three years.
What programs do you have in place to support that?
Samueli: We have internal R&D efforts in the 60GHz space. We have products that are being developed.
And you expect to bring those to market in the next couple years?
Samueli: Yes, definitely.
You've been making communications chips and radio chips, but there are a lot of applications processors and CPUs out there, too. What are you doing about application processors on the mobile phone, as long as you've got the sales pitch to customers of one big package of integrated chips?
Samueli: Today we're one of the largest shippers of application processors in the world. The difference is if you look at the span of applications processors, you have a very thin slice at the very high end, and then you have a lot of mainstream 1.5GHz-class processors which form the bulk of the shipment. Then you have some low-end, sub-gigahertz CPUs. The market we've chosen to address is the mainstream high-volume segment. Virtually all of our cell phone SOCs [system-on-a-chip processors, which integrate multiple functions into one package] that we ship today contain either dual-core or quad-core ARM CPUs running at 1.2GHz to 1.5GHz. At the very high end, the 2GHz-plus segment...
Like the Qualcomm Snapdragon 800s.
Samueli: Yeah. That's a much lower-volume segment. Over time, we will extend our portfolio to address those high-end segments. But we've made the conscious decision to go where the volume is today.
What are your forecasts for the best way to move beyond Moore's Law, to keep the computer performance going? As you get to ever-smaller transistors, you run into a barrier. What do you do beyond that -- photonics, spintronics, III-V materials? What's the next step after traditional silicon CMOS?
Samueli: Your guess is as good as mine. Nobody knows. That's the problem. CMOS has been on an amazing run for 50 years, and to just snap your fingers and expect another technology to supplant it and continue that run for another 50 years is optimistic thinking. My personal view is I think it's going to be a slog. We're going to push Moore's Law as hard as we can. We'll stretch it for another 15 years or so, but nobody has yet come up with a viable technology to extend the transistor further. We may really run into a dead end there where you may not see this kind of scaling over the next 50 years like we've seen in the last 50 years. We really are going to have fundamental physical limits slowing down the progression of computer performance.
We already have seen some of those limits -- not Moore's Law, strictly speaking, which just governs how many transistors you can lay down on a silicon chip, but in terms of limits when you're trying to get performance out of them. Power constraints have pushed chipmakers toward very multicore designs. Do you think software can catch up to the multicore world, or is that fundamentally an intractable problem for most programmers?
Samueli: I don't think that's going to be an insurmountable challenge. Anytime you put a challenge out there, people come up with a creative solution on the software side. I agree the multicore approach is the way to deal with the power dissipation limits you run into with CPUs, so you will see many more mainstream, multicore chips. Even cell phones have quad-core processors. People are figuring out how [to use] very efficient compilers and load balancers that distribute the processing across multiple cores.
You announced ARMv8 server processors [with technology licensed from ARM Holdings]. Does ARM have a shot at the server market? You're looking at just network-specific functions, but what's your forecast -- is there room to take on Intel in the server market?
Samueli: If you look at where our product portfolio is, we focus more on the communications processor segment of the market as opposed to the pure server segment. It's still an infrastructure market, but it's an infrastructure that's performing communications processing like deep packet inspection and security processing. There today we're shipping MIPS-based processors -- we have a portfolio of MIPS-64 custom CPUs that we've designed. And recently we've announced an ARMv8 version of a custom architecture that we'll be developing. When we introduce that product, we'll have the world's highest-performance ARM CPU. We talked about a 3GHz-class CPU in 16-nanometer CMOS [complementary metal oxide semiconductor manufacturing technology]. That will target the communications processing segment because that's where we have the best foothold and ability to penetrate the market and have done well so far with our acquisition of NetLogic.
Can some of those products apply to the server segment? They can, and maybe over time we'll look at expanding into that segment. It's to be determined whether people will be successful in the server segment against ARM.