Wednesday, August 13, 2008

FPGA vs. ASIC and FPGA trends

This semicon blog is a continuation of the recent discussion I had with Vincent Ratford, Senior Vice President, Solutions Development Group, Xilinx.

FPGAs vs. ASICs
Traditionally, there has always been a classic debate: FPGAs vs. ASICs! To find out more on the current status of this ever going discussion, I asked Ratford for his thoughts on this topic.

He said: "To answer this you have to take a long term view and look forward five years to what a 22nm device might look like and the cost to develop it. ASIC starts have declined over the last five years and continue to decline with ASSPs taking their place. For ASSPs to continue, they have to amortize the development cost across multiple customers, but with $50M development costs climbing to $100M, the number of applications that can support this is shrinking.

"FPGAs now look like SoCs with embedded processors, signal processing, multi-gigabit transceivers and a broad portfolio of IPs available from Xilinx or third parties. With shorter design cycles and increasing need to differentiate or die, we believe the world is increasingly turning to programmable devices.

"The investments we [Xilinx] are making in IP and software to enable more complex systems to be designed will carry our silicon platform forward and enable growth. Our challenges, going forward, are reducing power, providing more capability at a lower cost and simplifying the programming. As we make progress on all these fronts, we will take share from the ASIC/ASSP providers."

Global FPGA industry trends
Given the current semiconductor scenario, there is a need to estimate the global FPGA industry. According to iSuppli, the programmable logic market was $3.6 billion in CY2007. Xilinx revenues were $1.841 billion.

Given the wobble in the current global semicon scenario, where is the FPGA industry headed? Ratford said that Xilinx is forecasting 5-9 percent growth in FY09, which started in April 2008.

He added: "Our most recent quarter ending June was a record quarter for the company at $488 million, with 4 percent sequential growth. Our customers' design cycles are 12-18 months, so that tends to delay any weakness. However, design activity has been and still is strong. Our customers are highly diversified with over 20,000 across all the vertical markets. We see strong growth in ISM, automotive, aerospace and defense, and slow growth in telecommunications, which is our largest market at ~46 percent of sales."

According to him, wireless telecom, which constitutes about 40 percent of Xilinx's telecom revenue, is starting to go through a round of infrastructure buildouts with LTE and TD-SCDMA. This occurs about every five years.

"As these start to deploy, we expect to see some growth beyond what we are currently seeing. Finally, customers do a lot of prototyping with our platforms. We see no significant slowdown at this time," he noted.

Taking on Altera
Altera recently released the Stratix IV FPGA and HardCopy ASICs. It would be interesting to find out whether Xilinx has released anything that is close to or better than the Altera Stratix IV FPGA or the HardCopy ASIC.

Xilinx's Ratford said: "Our Virtex-5 products on 65nm have been shipping for 18 months, growing quickly and gaining market share vs. Stratix-3, which is also on 65nm. Since Altera has had difficulty executing on 65nm as far as Stratix III is concerned, they've had to jump to 45/40nm and recently pre-announced Stratix IV. Altera is saying they will ship first samples by December, however it takes quite some time to ramp to volume on a node and to put in place all the software and IP required.

"Our 45/40nm design has been underway for some time and we'll be in the market with a complete solution at the same time as Altera. What matters is when you ramp to production and when you have critical mass of IP so customers can start designs. We have a much larger IP portolio. Customer design activity on high-end FPGA will remain on 65nm devices for sometime to come. We estimate that Xilinx has about a 97 percent market-share in the high-end with our Virtex-5 family.

Alternative to HardCopy
Apparently, Xilinx also has an alternative to Altera's HardCopy. Ratford pointed to Xilinx's EasyPath, which provides similar cost reduction path for customers.

"Our approach doesn't require design compromises like that of HardCopy. Altera's most recent quarter showed HardCopy revenue down 2 percent to about $10M/quarter. It's insignificant!"

The last part of this discussion continues later this week! Watch this space, dear readers.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.