Texas Instruments CTO Ahmad Bahai discusses trends in analog for 2013 and beyond with Joe Desposito.
JD: The trend toward integrating more analog functions on a single chip seems to be gaining momentum. What might be the endpoint of this integration trend?
AB: The answer really depends on what kind of analog integration you’re referring to and the purpose. Different road maps and dynamics exist depending on the situation.
If you’re talking about monolithic integration, that’s one thing, and it has its own dynamics. If you’re talking about SIP (system-in-package), package-level, or module-level integration, that’s a different story. And again, the answer depends on the purpose of the integration.
In some cases, we are incentivized or encouraged to integrate for ease of use. Texas Instruments offers many integrated products along that line. In other cases, integration is performance-driven.
For example, a designer may need an integrated power amplifier with matching circuits to avoid RF noise. Sometimes integration is cost-driven, which is mainly for high-volume applications.
Another aspect of integration is development time and cost. An SoC (system-on-chip) development containing high-performance analog blocks needs a large multi-disciplined team and in some cases multiple iterations in a process that is not necessarily well-modeled or tuned for analog.
In contrast, a SIP approach, where the functions are directed specifically to a digital or analog chip on an optimized process and perhaps to a MEMS (microelectromechanical-systems) sensor element, minimizes both cost and risk. It is also easier to bring out a family of products given the multi-element flexibility
On ease of use, we see demand for more integration, mainly to reduce time-to-market. Our customers, who are not really IC designers, would like to see a more integrated solution for a broad range of applications, bringing together passives, multiple drivers, and data converters.
In many cases, this does not necessarily mean monolithic integration. It’s module integration where the packaging has the look and feel of a monolithic solution. We see this gaining momentum.
More companies would like to have what I call kind of a “plug-and-play” solution. This can really reduce the headaches and need for expertise in analog board designs. And that’s very exciting for us, too, because we feel that it gives us a chance to offer more of a solution than just the component and build up more momentum working with customers over the long run.
For performance-driven applications, integration should be applied more selectively. If you look at communications infrastructure, for example, such as high-end, big basestations, you see that designers still want high-performance devices, and sometimes integration and performance are a careful tradeoff. They want gigasample analog-to-digital converters (ADCs), high-performance amplifiers, and, of course, very low-noise amplifiers and such, and that will continue.
When it comes to slightly larger volume pico or small basestations, integration makes a lot of sense. Performance is, of course, important, but it doesn’t have to be at the level of big basestations that require standalone high-performance devices. That’s why we have a range of products in development that integrate the entire front end of basestations, but mainly target the smaller basestations. These products will address everything after the low-noise amplifier (LNA), all the way to the data converter and, in fact, we have the back-end DSP for all modem functionality.
So, all you need are two chips after the LNA to design the whole basestation plus power management. That’s integration driven by performance. Depending on the level of performance, you’ll find different levels of integration.
The last category, cost-driven integration, is really for high-volume applications, such as mobile. In essence, this category is dominated by applications with very fast time-to-market and is very form factor-sensitive like mobile consumer products. This is where integration has a totally different angle.
In some cases, the applications are dominated by digital chip requirements. In other cases, applications are dominated by the fact that there are only a few major customers who are interested in having the fewest number of ICs in the final product so they can design products more quickly. Time-to-market cycles are faster because, as you know, things get really outdated in the mobile market very quickly.
This category is building up momentum. You can see IC designers integrating many dc-dc and LDO (low-dropout) converters in a single chip, and that trend is going to continue. PMICs (power-management ICs) are becoming more and more complex. In fact, PMICs at this point are quite complicated considering that they have battery management, lots of LDOs, switched mode dc-dc converters, battery monitoring, and, in some cases, chargers.
All this on one chip—a giant analog power management chip. In a cost-driven chip, performance is often compromised to realize all functions on a relatively low-cost process that has few special analog layers.
All of these integration roadmaps are happening, but depending on the category, they have different paths and momentum. For cost-driven integration, we see the biggest push. The performance-driven is not as strong. And in the broad market, we see a definite incentive for engineers to have an easy-to-use module. But of course, as I said, some of the broad markets use package-driven or SIP-driven integration. This is not affordable in high-volume mobile markets, but in some broad markets it’s highly justified.
JD: Do general-purpose chips have a future? Or will every analog chip become part of a solution for a particular market?
AB: At TI, we have strong programs in place for medical, audio, automotive, and other markets, but especially automotive. Automotive is very hot for all semiconductor companies.
In analog and mixed signal, and to some extent in digital, you really need to have a portfolio of high-performance IP blocks and ICs to build up an application-specific standard product (ASSP) presence. One of the unique strengths of TI is that we have more than 40,000 analog products, and that gives us huge leverage to put products together to address a specific need for a specific market.
So, from a technical standpoint, you need high-performance IP blocks to have a strong ASSP presence in any automotive, medical, or high-performance audio market. And I don’t think we can bypass developing high-performance standard linear products. We still need high-performance amplifiers and data converters and clocks and such to address some of these larger chips.
So, no, high-performance standalone products are not going away. They will always be there. From a business and marketing standpoint, though, there are markets and applications that are really at the highest end of performance. They still demand standalone, standard linear, high-performance analog blocks sometimes.
In some cases, the volume is not large enough to justify integration. But when we deal with a large volume market like mobile, we don’t hesitate to invest in integration because if you are successful in that market, it’s going to be hundreds of millions of dollars.
Let’s consider medical. You need performance for MRIs and imaging like in the real-time X-ray machines that some of our customers build. It’s a good business for TI and it’s a healthy business. But the volume is not so huge that it justifies constant integration. Add to this the fact that performance is still at a premium. We cannot compromise performance for cost when it comes to real-time X-rays and a catheter into a brain, for example. So, in those cases, standalone high-performance analog will dominate for as far as I can see.
It makes sense from a technical viewpoint and from a business viewpoint. These machines are in hospitals, not in a doctor’s office, so the volume doesn’t justify too much integration. We need to continue to build up high-performance standalone products and also ASSPs. There are markets that really need performance at the premium price without too much integration.
JD: Gallium-nitride (GaN) and silicon-carbide (SiC) ICs are gaining more traction in high-voltage applications. Does TI have any plans to compete in this area?
AB: High-voltage parts have a lot of room for growth in the semiconductor business. The high-voltage area has been served by great traditional technologies like IGBTs and also technologies like superjunction, which is more recent. But in terms of power density and in some applications where you need higher switching frequencies, the road is open for a lot of innovation.
And one thing about GaN and SiC or in general III-V material that makes them quite exciting is that they combine three main “like qualities” that are hard to find in any other material. These devices can handle high voltage because of their wide band gap nature. They can handle higher switching frequencies because of their higher electron mobility, and they can handle higher temperature. So, this combination makes them very exciting for offline applications, motor drives, or data centers. So, absolutely, we have a lot of interest in this, and we are involved in these programs in many different ways.
JD: Do you see any other trends that we should be looking for in 2013 and maybe a short time beyond that?
AB: Yes. You know that famous saying, “It’s hard to forecast the future unless you create it.” Well, I’m a big believer in that one, and at TI, we are well-positioned to play a big role in creating the future. So, I want to emphasize three areas.
One is the miniaturization of PMICs, which have tended to fall behind signal path ICs in analog for many years. This will build in momentum, and you will see an amazing improvement in the power density of PMICs, which is critical for almost every market we just discussed. I think you’ll see a lot more of this in 2013. So, power management is going to improve—not as fast as Moore’s law—but it’s going to pick up a lot of steam in terms of moving toward a much smaller footprint with higher power density.
I also see an exciting opportunity with MEMS. Although MEMS are mature in many markets, including consumer and infrastructure, these devices still demand some specific requirements of the analog front ends (AFEs) and integration of them. I expect to see a lot of interesting chips coming out in MEMS-related markets integrated with AFEs.
Also, one derivative of that, which I see picking up, is the intelligent sensor market. I’m talking about embedded sensors such as MEMS plus AFE plus microcontrollers. It has been a hot market, but hot only on the technology side. I call these markets “zero billion dollar markets.” By this I mean that a market is very exciting but nobody is making a billion dollars from it. We think that it’s time for embedded sensors to move to the next level. Some people refer to this as the “Internet of things,” but I would emphasize embedded MEMS containing analog and digital circuits.
JD: What are the trends you’re seeing in the digital realm?
AB: All the processors, application processors, and DSPs are shifting toward multicore architectures. And then, of course, Moore’s law is at the point where you cannot just rely on scaling. You need to look at new materials and devices. In fact, some major digital companies are applying more exotic III-V materials to CMOS technology to increase the speed and reduce the leakage of these devices when they go below 22 nm.
But what I would like to address is the analog part of digital chips. Digital chips are becoming much denser. We’re talking about millions of transistors on these chips. Because of this, power management of these multicore chips is a lot trickier than it used to be, in the sense that you need to tie the power management and processor very tightly together, so that it can react to the power requirements of each core almost independently, and, of course, help in managing the leakage and all that.
The co-integration of power management with these big processors is becoming more and more of a necessity rather than a “cool” factor. The convergence of power management and large digital chips and multicore is an area that is a nice intersection for analog power management, as well as the future of the more advanced big processor chips. There are a lot of obvious device innovations, such as FinFET and all these technologies that designers are using to improve chip performance and also embedded memories, but that is a long story in itself.
JD: Should we still be making a distinction between analog and digital ICs, or are chips that combine both the new normal?
AB: That’s a good question, but it can have several meanings, and depending on what you mean you have different answers.
If you mean that in most SoCs there is a lot of analog and digital coexistence, single chip or single module, yes, that’s happening and there is going to be more of that, obviously. So, that’s more like a coexistence of analog and digital on a single die or package. It’s been around and there will be more of that. It’s going to be here for as long as you can see.
The second element is that analog needs to intimately work with digital to improve digital performance. It’s not just analog and digital coexisting or that they have to coexist because of integration. It’s that we can fix some of the analog difficulties with digital or make things more programmable or improve the linearity or things, and that is the second level of coexistence, which I call “intimate coexistence.”
The third element, really the trend of the future, is that when digital gets to the point where you are transmitting, transferring, or interfacing with gigahertz or gigabit signals, even if it is deemed a digital signal, it really is an analog signal, because clean ones and zeros no longer exist.
Look at LVDS (low-voltage differential signaling), for example. It’s supposed to be a digital interface, but it is really an analog signal because the ones and zeros are totally distorted after transfer and transmitting over a copper wire, a PCB (prointed-circuit board), or even inside the chips.
So when it gets to those frequencies and that level of performance, digital signals look very much like analog. When you’re dealing with them, you’re no longer dealing with ones and zeros. You need to do a lot of pre- and post-processing before you get to that one and zero level. So, from that viewpoint, there’s even more convergence. There are really three different elements, and each has its own momentum and dynamics.
JD: Engineers are always interested in datasheets and app notes. But with analog integration and other things changing, datasheets aren’t the information source they used to be. How is the datasheet’s role changing?
AB: What we see from our customers is that more and more of them really would like to see the complete solution rather than just a biography or datasheet of each component. In other words, they want to see how to solve their problem.
If you just offer them a data converter, for example, and give them a typical number for ENOBs (effective number of bits) and SFDRs (spurious free dynamic ranges), and things like that, that’s a great starting point. But then, how do you put it together with a driver and clock and deal with all the noise and the issues that you have around that. In other words, how do I get this thing going?
At TI, we try to offer a complete solution to customers to make life easier for them. When it gets to higher performance, though, putting the parts together in itself is a big challenge. Of course, one aspect is further integration, but otherwise, we need to offer a set of ICs that can come together a lot more fluidly. Struggling through a lot of design work to make the parts work together, especially if they’re from multiple vendors, can be a nightmare.
Customers are not asking for just integrated solutions, but complete solutions composed of multiple parts that can come together in a user friendly manner. And that trend is building up quite a bit. In TI as the leading analog company, we offer many design tools such as the WEBENCH design environment by which customers exercise the solution without having to wade through a set of thick datasheets and application notes.
JD: Some of these datasheets have become 300 pages long because they cover every part of either an integrated chip or maybe an integrated solution, so engineers feel like, “Well, let me talk to some other engineers and see what they think.”
AB: Yes, that’s right. The more complex these chips get, the more complex the whole design becomes. For us, as a semiconductor company, it’s more advantageous to offer complete solutions because this makes life easier for our customer and makes the design efforts easier and, hopefully, longer lasting.
Ahmad Bahai is chief technology officer of analog business for Texas Instruments and director of Kilby Labs and TI Silicon Valley Labs. He co-invented multi-carrier spread spectrum theory, which is used in many modern communication systems such as 4G and power-line communications. He received his MS from Imperial College, University of London, and PhD from the University of California at Berkeley, all in electrical engineering.