Company: Xilinx
Established: 1984
Headquartered: San Jose, California, USA
Employees: 3,300
Business: Programmable Logic Devices & Solutions CEO: Moshe Gavrielov
Revenue: $1.84 billion (FY 2008)

 

The Spartan 3AN Flash/SRAM hybrid FPGA device from Xilinx

 

 

The way that Xilinx approaches the market seems to have changed quite considerably. In the past it was just a matter of selling the FPGA (field programmable gate array) devices required, now with embedded processor and DSP (digital signal processor) IP cores being utilised, as well as more complex development tools being involved, it is more of a combination of the silicon and the software elements. Would you now class yourself as a solutions provider rather than a simply a semiconductor firm? Well, I think we are in the middle of what you might call a period of transition at the moment. If you look at what FPGAs could do ten years ago and what they are capable of doing today, the difference is quite striking, I would say far more than any other semiconductor sector. There are quite incredible levels of integration involved in these devices now, and they can be optimised for a range of different design requirements; such as signal processing, memory, logic, I/O, etc. Because of that they need a high degree of software support to go with them. I guess that it would be fair to conclude that we are becoming a provider of solutions rather than just another silicon firm. That means that we are now more at the centre of our customers design requirements rather than a peripheral part. How has this effected your annual research and development spend; what proportion goes onto each of these? Currently just under 19% of our overall revenue goes back into product development, which is considerably more than most semiconductor firms can claim. This equates to around $350 million per annum. We break this up fairly evenly between the three main sectors of our product offering, with a third being invested into silicon, a third on IP, and a third on software. Is there a risk that the expanding suite of design and development tools you produce will encroach on those of your EDA (electronic design automation) partners? How do you make sure you don't cross that line, and end up being in competition with them, but still adequately support your customers? I personally think that there is still a lot of room for EDA vendors to produce design tools for the programmable logic sector, and I don't see any reason why there should be any serious overlap between what we offer to support our chips and what they do. We are only really concerned with the silicon implementation and the place & route aspects, but as you move toward the front end of the design it is the EDA firms that have all the key competencies here, and I expect that to remain the case. There are opportunities for these vendors to produce more comprehensive software tools for the FPGA market, as a greater proportion of system designs begin to utilise these devices. They will look to concentrate more of the efforts here as programmable logic gains in popularity due to the economic and timescale concerns. The budgets for programmable logic-based designs are quite different to those of the ASIC sector though, so the business models that EDA firms employ will have to reflect that and they will need to find more innovative approaches in order to produce strong revenue streams here. Xilinx has to some degree outgrown the traditional FPGA market of the past, and has been pushing in recent years, with the introduction of ASMBL-based FPGAs, to take a sizeable bite out of the far bigger ASIC (application specific integrated circiut) market. Finally when it looks like this is beginning to happen the company seems to be streamlining its operations and shedding staff. Isn't there a risk that this will limit your effectiveness in these new areas? The shedding of the staff is more the result of the company going from a business unit structure to a functional structure. Even though from the outside it may look a little counter intuitive, thanks to this reorganisation we will be able to address these markets more efficiently as we will no longer have a fragmented organisation with each unit only looking after its own interests. That will mean we are far better placed to follow an aggressive strategy that will allow us to grow our bottom line. We are just simplifying our logistical and administrative overheads, but we certainly aren't under-investing in the sales or technology sides. Far from it. With the move to lower node sizes isn't your development of new chips basically dictated by the speed at which your foundry partners can implement new processes? Would you say that Altera has managed to get the jump on you with the release of a 40nm FPGA thanks to its alliance with TSMC (Taiwan Semiconductor Manufacturing Company)? It's a very valid question. We have managed to gain huge market shares at both 90 and 65nm, and we want to capitalise on this by offering top quality products at the next process node. We are putting a great deal of our resources into our next generation of FPGA products, and plan to make announcements on this matter at the end of our fiscal year, in March 2009. Is this something of an embarrassment for Xilinx, as market leader, and with a track record of being first to market with programmable logic devices that utilise the next process node? Pre-announcing products it easy, but the question is when will a stable, fully tested, solution that clients can actually use be available. We have gained a reputation for supporting our customers really well, and we wouldn't want to jeopardise it with publicity stunts. Our competition has not been able to gain a notable market share at the previous nodes, and we don't expect them to get the lead over us here either. We are working diligently to produce the software and the silicon that will give us a compelling FPGA solution at the next process node, and this will allow us to gain the same sort of market share as we have seen at 90 and 65nm. Will Xilinx release lower node versions of its Virtex and Spartan devices in close succession, or is the need to produce lower-end products at this node not as great? We think there is a need for both sides to be covered. We plan to have next-generation FPGAs that address a broad portfolio of market needs, so there will be high end and cost sensitive solutions being announced by Xilinx in the timeframe we have already discussed. The consumer sector has been a target for Xilinx for some time, but despite the advances made in new FPGA and CPLD (complex programmable logic devices) architectures, aren't the same limitations; power consumption, chip size, unit cost, still there? Can the flexibility and time to market advantages of programmable logic ever really counterbalance these downsides to a large enough degree? Will your in roads into this sector always be restricted to certain high-end, relatively small volume products like the larger flat panel displays? Your observation is correct, if you look back to 2001, Xilinx was relying very strongly on the communications infrastructure sector, with this making up around 80% of its overall business. Today we have a much better balance, with 40% coming from communications, and the rest made up of automotive, industrial consumer, military, etc. I agree that in order to move more aggressively into the consumer space, the issues of power consumption and cost need to be tackled. Though these are huge challenges, headway is already being made. Once they have been overcome, it will mean that a larger proportion of system designs can be done on a programmable fabric. If you look at the economic paradigm shifts that are taking place currently, with much shorter design times and the increasingly fickle nature of the consumer sector, the benefits of using chip solutions that have programmability, are quick to produce, and don't require large upfront development costs are clear. FPGAs are now becoming a more viable option here. I am convinced that this market will open up to us the same way other markets have opened up. The high volume, stable designs like gaming consoles are out of our reach for the moment, but there are lots of areas where FPGAs are gaining ground. I have worked in both the ASSP and ASIC sectors in the past, and the reason why I joined Xilinx is that I think that programmable logic is the way forward. I see enormous potential for growth here. About 18 months ago Xilinx introduced its first flash-based FPGA solution, in order to address the high-rel market. How much of an uptake have you seen on this, have you taken some business off the incumbent supplier in this arena Actel, and do you expect to continue releasing products using non-volatile technology? Yes the Spartan 3AN device has seen strong adoption in sectors where high-reliability or security issues need to be taken into consideration. It provides a highly integrated solution, with the respective benefits of both SRAM and flash elements on a single device, and there may well be other areas like this that can be exploited by using some form of multi-die technology. Moving forward, we'll look to pursue the opportunities that offer the highest return on our investment. There are so many avenues open to us currently that we need to make sure we address the ones that are the best prospects for our long term growth. When Xilinx had a business unit structure, each of these having a lot of autonomy, they might have made decisions that were good for that particular unit, but not particularly beneficial to the company as a whole. Now we have to look to fight bigger battles elsewhere, and use all our available resources, not just go after relatively small bits of new business left remaining within the traditional programmable logic sector. Taking a bigger chunk out of the ASIC business is our top priority. Standard cell solutions are now really struggling to meet the needs of the market, so this gives us the chance to take advantage of their shortcomings. The wind is clearly in the sails of programmable logic, while ASICs are sailing against the wind.