Tensilica and Customizable Processors

Introduction

According to Wikipedia the name Tensilica comes from a combination of the word tensile, meaning capable of being extended, and the word silicon.

Tensilica has two main product lines of 32-bit processor cores for SOC design: the Diamond Standard processors and the configurable and extensible Xtensa processors. The Diamond Standard processors are a set of off-the-shelf synthesizable cores that range from area-efficient, low-power controllers to an audio processor, a high-performance DSP, and a video processor. Xtensa processors are synthesizable processors that are configurable and extensible. Using Xtensa technology, the system designer molds the processor to fit the application by selecting and configuring predefined elements of the architecture. Additionally, designers can optimize the processor by inventing completely new instructions and hardware execution units that can deliver performance levels orders of magnitude faster than alternative solutions.

In interviewing EDA personel a common theme is that it is fairly easy to sell a solution that is plug compatible with a prospect’s existing design flow and that offers demonstrable and substantial improvements in execution speed, capacity and/or quality of results. A simulator offering 5x speed improvement would be a classic example. An improvement of 2X in speed is generally insufficient to generate much excitement. It is much more difficult to sell a product that requires a change in design flow or methodology regardless of the benefits that may be claimed by the vendor. It is also difficlut to prove to a prospect that they can achive the tauted benefits. It is not as simple as running a benchmark. In the case of Tensilica it is eaiser to convince prospects to purchase the off-the-shelf Diamond Standard processors rather than the configurable and extensible Xtensa processors. In fact the former are used to gain credibility with prospects and help persuade the prospects to purchase the more capable Xtensa processors.

I had an opportunity to discuss the situation with Jack Guedj, the newly (July 16th) appointed CEO at Tensilica

Jack, this is Jack.
Jack, this is Jack
This is going to be confusing.

How do you pronounce your last name (Guedj)?
Gage. It is interesting. Over the weekend I saw someone who has the same type of name but spells it Gaige. I am guessing some other people have the same type of name but it is just a case of different spellings.

Is your real name Jack?
That’s my middle name.

My name (Jack) is a nickname. A lot of people called Jack are really Johns.
Mine is Jacques. I am originally French.

I noticed that on the Tensilica website. Perhaps we could begin with a brief biography.
Sure. Before Tensilica I was running Magnum Semiconductor. Magnum is a spin out from Cirrus Logic. I started with Cirrus and then we spun Magnum out, raised $24 million and took the group private out of Cirrus which is still focusing on mixed signal products. There are no more digital products at Cirrus. Then last year (June or July 2007) we acquired the consumer products group of LSI Logic which was largely the old C3 company that LSI acquired a few years back. Prior to Magnum and Cirrus I was with TVIA, a company making streaming video processors for broadband set-top boxes and digital TV. That was the era of the Internet everywhere. But at the time the Internet was too slow. So it was hard for consumers to use that in their living rooms. We were into set-top boxes for GI and Microsoft. GI which is now Motorola; Microsoft set-top boxes, MSTV, as well as a number of digital TVs like Sony and Mitsubishi. We took Magnum public in August 2000. This was one of the last company to go public at that time. We raised $65 million in an IPO which was five times oversubscribed.

Prior to TVIA I was with Faroudja and National Semiconductor. I don’t know if you are familiar with Faroudja. They were making all the high end video boxes. At that time most of the world was in standard definition. Those boxes were converting standard definition into high definition. Faroudja is now part of Genesis Microchip Inc in their display processor division.

At National I was directing residential broadband and consumer market segments. There was a very interesting thing that Gil Amelio did after he put National back on its feet financially. National had been losing tons of money. He stabilized the company. National had a bunch of product lines. They were all operating, doing what they should do which is to focus on their own product lines and their own market segments but there were too many of them. So he created a superstructure on top of the product lines to create business plans that would tap the different product expertise the company had. The only way he thought about changing their behavior was to drive their R&D investment based out of this strategic market segments, to just pick a few of them and make a total play. That was very interesting because we recovered some relationships. At Motorola it took six to nine months to do that because of past bad history. We were able to install Sub23 which National had and TI was buying 270 million Sub23 at that time. That was one avenue. It turned out that several designs for other applications that had been done were a perfect fit for that set-top box. We had RF products, memory products, specialized products, .. It was really working well as an overall system play.

Prior to National most of my career was more involved with broadband communication starting with TI where I was involved with the broadband communication sector. That was in France. Also at TRW with RF devices where we were doing track amplifiers for cable TV as well as power amplifiers. We did the first TV airflow system Gigabit Logic which merged with Triquint going Gallium Arsenide and high speed communication. At that time I was fairly involved with the then fledging fiber channel SSCI channel standard.

You have advanced degrees in electronics from the University of Paris. How, when and why did you transition from the technical side to the operational side?
I have an engineering degree. I just saw a movie which portrayed the lives of Pierre and Marie Curie. The engineering school where I studied was where radioactivity was discovered. It was pretty interesting. I did a doctorate at the University of Paris. But I would say that I had an eye to be a generalist in business rather than having a technical career. While most people selected the school of Physics and Chemistry because of its very good research program, I selected it because it gave me an overview of many fields. I thought that it would help me in the business world later on. If you look at semiconductors, you have these fields including software, semiconductor design and semiconductor physics and processors. If you can span your knowledge at a broad level, I thought that it could help me. While studying physics and chemistry I started a club as part of the National Chamber of Finance. We started with a couple of people and grew that to close to a hundred people. This was a funnel for the national association. They were able to have people who would join that organization later on. That was done in conjunction with a lot of business schools in France.

One more thing. The reason I came over to the US, what basically drove me to take a one way ticket was to get my MBA at UCLA was because a lot of business theories come from US schools. That is where I wanted to get first hand information.

Editor: Some people know from the moment of their conception what profession or career they wish to pursue. A fraction of these people achieve their goal. Many simply change their mind over time. Changing majors and therefore career choices is a well established college tradition. Some are impacted by personal issues such as health or finance that disqualify them or make it very difficult to follow their desired career path. Some are impacted by external circumstances. Many an aeronautical engineer was forced to abondon their chosen profession when diaster befell NASA. The economy can wreek havoc on a career.

Jack Guedj’s planned to be a business executive in high tech and choose what for an American would be a highly unusual path towards that end, namely a doctorate in electronics by studying physics and chemistry at a prestidous French engineering school before earning an MBA at UCLA.

What attracted you to Tensilica?
First of all it is a company that has succeeded in the IP business. A lot of companies have had difficulty in the IP business. To get that kind of momentum is a testament to how good the technology is and the products are. Virtually everyone I talked to including potential competitors were telling me that they had a very good engineering team and very good products. This is a good foundation. And by the way so far since I have been with the company my observation is still the same. That is what I got from the outside before I joined the company and what I am getting now from customers that are using the product is the same. They are very pleased with it and pleased with the support our engineers are providing. It was an interesting challenge to basically take that solid base in terms of both product and revenue and be able to take that to the next level; to make the footprint two or three times the current size.

How does Magnum Semiconductor compare to Tensilica in terms of size, business model, stage, …?
Magnum is over three times the size. Magnum is an SOC supplier. It is a customer or could be customer of the type of product Tensilica has. It is on the other side of the fence from Tensilica. Consequently, I have first hand appreciation of what a customer would want to see and why and of what would matter to them. That is pretty helpful. As it turns out I have a lot of colleagues that are in the semiconductor industry and that are Tensilica customers today. It is interesting form the customer’s side and I view this as an advantage that I have a lot of friends that are or could be Tensilica customers.

According to the press announcement your objective over the next year or so is to lead the company in its next stage of growth.
We are reviewing where we are with current products. It is actually shorter than a year. I wish I had a year. The tech industry moves very fast. But absolutely to look at the direction that the company is taking and where is the best bet for the company. In some cases you need to narrow the funnel and focus in order to get the maximum impact. By doing so you are taking a greater risk, that’s the name of the game.

Is the key to growth investment in sales and marketing, new products, new territories, ?
Of course it is a combination of everything. I think we have tremendous opportunity in Japan. If we had more sales persons, we could grow faster. New products? Absolutely! The more products, the more perfect the product is, the more features the better off you are. But largely I think a more marketing and sales orientation. Based on that engineering will follow. If I had to look at all the issues on the table, I would give a push to sales and marketing.

What is the biggest challenge or obstacle to achieving the objectives?
The company has made significant effort in developing standard cores and standard products in addition to its foundation which is Xtensa customizable processors. That is an area that is pretty well developed, video and baseband communication. Those three areas have shown that the company adds value to first to customers who do not like to get into the details and just want a ready-made solution and second to customers who can not relate to how customizable processor technology will help them. Say someone is in baseband communications. Based upon the understanding that comes from some benchmark on how well we can do for some kind of error correction circuitry, they realize what they could do with the product in addition to having programmability. The standard product strategy has helped us twofold. One for the customers who want a ready-made solution but also for the potential customer of customizable processors that now can relate to how they can solve their problems with the products that we are bringing. That is our challenge. In many cases if you look at it once the customer uses the product the feedback is very good. I got feedback from a top five semiconductor company saying that “We are using your solution for audio. We tried to come up with algorithms that we could not run. We came up with a list of them but so far we have been able to do them all.” Once customers use the product, they are really happy with it. The challenge for us is to get more connection with the customers so that can lead to “Okay, now I get what this product can do.”

How big a company is Tensilica?
Revenue?

Revenue, employees, ../
We are about 135 people. Current revenue is in the high twenties.

Tensilica has two primary product lines: Diamond is the off-the-shelf- standard products and Xtensa is the customizable processor line.
That’s the current product names. I can share with you that we are moving toward a functional segmentation. As you know HiFi is what we do for audio. We are moving towards that for audio, VDO for video and Vectra for baseband DSP.

Is the growth potential for Tensilica in standard processor or customizable processors?
I view the problem more as what can we do for someone doing audio, for someone doing video to solve their problem. We need to have the expertise. We need to be viewed as the experts in those areas. On the audio side we support forty to fifty different standards. When we walk into the customers’ shop, we offer a full solution. Yes, we have some customers who have specific requirements. We can seamlessly enable those customers to use standard cores and to go to the next level which is standard cores plus predetermined configuration. That is very simple because you go on the software and just click on the configuration type. That is all you have to do. If you want to further optimize, then they have the possibility of customizing that. It could be just for audio or mixing it with some additional functions, or taking a subset. Then they have these customization options or they can simply write some C-code or Verilog code type formats instructions that they can customize. The big part of what we do is not only enabling them to come up with a customized processor but really what we offer and spend a lot of time and effort on is finding out the whole set of software tools that come along with that specific processor, i.e. compliers, debuggers, verification tools. If you look at product development, we are going to introduce two new major products at the end of this year. Basically one year after we have delivered a core to a couple of alpha site customers, we make sure for all the generations of the customized versions that all the tools and systems are solid. We are spending a lot of money enabling us to seamlessly provide an entire suite of software to the customers.

What is the value pitch to the potential customer? What does this approach provide that the customers’ current tools and methodology does not?
Processors are good when programmability is needed. Programmability can be required in two ways. One is that you have a very fast evolving world. So you need something that is flexible enough so that if there is a new standard or an evolving standard, you need programmability to sustain that rate of change. Second you have certain functions that are continuously evolving. Take a video system. There are always ways to improve the picture. It could be tiny improvements but you have to be able to do that continuously over time. Therefore a programmable solution helps you because you do not have to spin another chip in order to get your improvements. Hopefully you did the right thing because if you did not, it could take a couple of tapeouts. Today mask sets are running $1 million at aggressive geometries. This is a very expensive proposition. Now once you consider those facts and say that I need something that is programmable, then we have the best solution because most processor companies or DSP companies will provide what they think is the best cut. They make one product and do all the testing on that product and provide a couple of configurations. Then you have to map your application to that. But nothing can be better especially when you start stressing performance than something that is customizable to your needs.

We marry processor, DSP, standard processor or what is called DLIW for very long instruction word. We marry all that and enable you to customize it at a click of the button to fit your needs. That is where we can get to performance levels and claim applications where standard processors would be very onerous in terms of silicon size and power.

Are there any competitors claiming to have similar capabilities in this area of customizable processors?
Good question. We are competing more with what is called RTL design. You can take an RTL design which is a fixed hardware configuration and build some register programmability, some flexibility. Some may think that they can get there. They may have to do some spins of silicon along the way. Again the challenge we have is to communicate better so that the customer can see the value upfront and not make the mistake where they need this programmability. But that approach is certainly entrenched. In some cases people have been using it for years. So it takes a long time for someone to relate to how you can really do that, to say “This is really interesting. Let me see how I can use the product for building programmability I want.” That’s the bulk of it. On the other hand you could take a generic processor and say once I have a generic processor I can do anything and try to stretch from that generic processor into some specific application processing. Those two are possible competition for us.

If you have an off-the-shelf product, presumably potential customers could simply benchmark it and persuade themselves that the product could reduce TTM, costs and so forth. In the case of a customizable processor, how do you persuade potential customers or how do they persuade themselves that your product can achieve a level of desired level of performance?
That’s a good question. That is absolutely the question. It goes back to what you and I were discussing earlier about the standard cores. Because the standard cores could do that, they could show that yes you can achieve this level of performance. You could look at libraries. In some cases it is good enough to say that we can do an FFT function or another function in some many microseconds. Or someone would say “Wow, you can do that!” Standard cores and libraries go a long way to give the customer the proper understanding of what the product can do

I noticed that you have offices in Japan, Beijing, Taiwan and South Korea. Where are Tensilica sales occurring?
I would say Asia and the US. The US is a big region for us. Silicon Valley is definitely good grounds, a lot of semiconductor companies. Asia is definitely a good territory for us. We have high hopes for Europe, an extended Europe including Israel. But the volume currently in Europe is smaller than either the US or Asia.

In Asia do you have direct representation or do you go through local third parties?
Mostly we do direct because we want to delight our customers. The products we are offering are complex enough that we can delight our customers more by being direct.

Do you see your end users moving in ways that will increase their demand for your products?
Thinking about it right now, what would you say is the biggest driver in the electronic industry?
Consumer products?

Consumer electronics. I would say the other one is the Internet. I think that people have forgotten that in 2000 it was going to happen. But now with broadband able to be available to consumers at the speed they can enjoy. They do not have to wait for the dial tone connection to happen. Those two things means that consumers will more and more receive content from places that they don’t know in formats that have multiplied. When you are talking about forty or fifty audio formats and I have not counted all the video formats but there are a ton of them out there. Whether you want to do standard formats on set-top boxes and start adding real video formats or VISIC formats or Quicktime or flash that you have on Google then the number of formats is increasing. So it is becoming increasingly difficult to do it efficiently for a hard-wired solution. The fact that the universe is increasing in terms of the formats. I include audio/video but if you are take a cell phone and go 3G and next generation which enable higher speeds which stresses a lot of circuitry as well as multiple formats. If you are in Korea, you want CDMA. If you are in Europe, you want GSM. All those multiple formats are driving the demand for Tensilica products.

In addition to that Moore’s law is working in our favor. With new geometry you can pack so many gates per square millimeter that the impact of programmability becomes very, very minor. The benefits of programmability outweigh the small impact on size. If you had one fixed format that is not going to change, you can always do it more efficiently with a custom design. But again the driving force through multiformats and the fact that geometries are such that you can pack so many gates in a small area is working in Tensilica’s favor. But the reverse argument of packing so many gates in a small area is that you have to build a lot of circuitry and functions in silicon to do justice and have a horde of designers working arduously to be able to put out all the custom circuitry on time for those accelerated lifecycles. The other item that plays into that which you mention is time to market. Lifecycles are becoming shorter. I used to have my TV for ten years. Now I have a TV for three or four years. All these factors are playing into Tensilica’s favor.

Is there an IPO in Tensilica near future?
I would oppose it. But no plans at this time. Only the plans to build a great company, a profitable company. I believe that the employees and investors can be very well rewarded with a great company. Beyond that there are options. If the company decides it is the appropriate time. There is the stock market side and you can not control that. You decide at the appropriate time what the best course of action is.

When you took over as CEO, Chris Rowen, the founder and pervious CEO, moved to CTO. Not an uncommon occurrence in high tech companies. How is the relationship and split of respo9nsibilites working out?
First of all Chris is the most knowledgeable about the Tensilica products. He is a unique person in that he can quickly grasp an application issue and come back either directly or through the help of some of the people at Tensilica about what is the best match, how it could be best addressed with Tensilica’s products or with a combination of Tensilca products and some existing circuitry that the company has. Part of this is not just a CEO issue but the fact that the company is getting bigger and needs additional strength in the executive team. So the company chose to do it this way. Chris will now have time to work with customers and map out the best direction in terms of current applications and future products with what the customer requirements are. I think that will be a very big portion of what he will be involved in. And the things we discussed about vertical segments. Chris will be intimately involved with the selection of those segments and in the development of the library and product directions for those segments.

I am out of questions. Anything you wish to add?
No! I am really excited about being on board and really pleased to see all those customers positive about Tensilica. When you join a company you worry abut what you were not told. So far I have not found any big surprise. On my end I have found more confirmation of the appreciation of how solid the products are.

The top articles over the last two weeks as determined by the number of readers were:

Altos Variety LX Adopted by STMicroelectronics for Ultra-Fast Corner and Statistical Characterization at 40nm Altos Design Automation Inc. announced that STMicroelectronics’ Data Storage Division has adopted Variety LX for corner and statistical cell characterization. ST will use Variety LX to characterize a commercial cell library augmented with its custom cells for use at 40nm. The characterized views will be used for both corner based and statistical timing analysis.

Cadence Reports Q2 Revenue of $329 Million Cadence Design Systems, Inc. reported second quarter 2008 revenue of $329 million, compared to revenue of $391 million reported for the same period in 2007. On a GAAP basis, Cadence recognized net income of $5 million, or $0.02 per share on a diluted basis, in the second quarter of 2008, compared to net income of $60 million, or $0.20 per share on a diluted basis, in the same period in 2007.
Although we achieved our Q2 numbers, it was more difficult than we planned. Customers are demanding still more flexibility in when, what and how they purchase software and hardware," said Mike Fister, chief executive officer. "As a result we've made the decision to lower our outlook and transition to an approximately ninety-percent ratable license mix.

Synopsys Announces Availability of New Fully Synthesizable PowerPC Cores Synopsys, Inc. announced the availability of fully synthesizable implementations of the IBM PowerPC 460 and cache configurable PowerPC 405 embedded microprocessor cores as components of the DesignWare Star IP program. Developed through a close collaboration between IBM and the Synopsys Professional Services and DesignWare IP teams, the foundry-independent processor cores are supported by a broad range of design tools in the Synopsys Galaxy Design and Discovery Verification platforms. The PowerPC 460S and 405S processors are distributed as simulation and timing models and synthesizable (RTL cores. Synthesizable IBM CoreConnect peripherals are also available to licensees of the PowerPC cores.

Synopsys is Selected by National Semiconductor as Their Key EDA Partner Synopsys announced that National Semiconductor Corp has selected Synopsys as their key EDA partner for use in product development. The expanded relationship provides National access to Synopsys' Analog/Mixed-Signal and custom IC design tools and the opportunity to collaborate in next generation solutions. As part of the agreement, National has chosen products across Synopsys' product portfolio

LogicVision Reports Second Quarter Financial Results LogicVision, Inc. announced its financial results for the second quarter of 2008, ended June 30, 2008. Revenues in the second quarter of 2008 were $3.0 million, compared with $3.0 million in the first quarter of 2008. Net loss in the second quarter of 2008 was $1.0 million, or $0.10 per share, compared with a net loss of $1.3 million, or $0.13 per share, reported in the first quarter of 2008.

Other EDA News

Other IP & SoC News

You can find the full EDACafe.com event calendar here.

To read more news, click here.

-- Jack Horgan, EDACafe.com Contributing Editor.