I recently watched a video interview with Umar Mughal, a senior manager from Altera, about where FPGAs will be in 10 years.
The video started with a review of FPGA applications over the last three decades, from debug through co-processing to today, when the device is at the heart of the system. These FPGAs were depicted as (probably) the most expensive item on the BOM, and I doubt many readers would argue with that. Today we have "FPGA" devices with predefined building blocks such as PCIe and processor subsystems. They are designed to give a further boost to capabilities, but they also blur the definition of where an FPGA begins and ends.
Many of the innovations Umar discussed are already happening, such as electronic system-level (ESL) design replacing RTL design. The trend toward higher levels of abstraction is clearly necessary if customer designs are ever to fill the rapidly increasing capabilities.
Having working in FPGA marketing for years, I have listened repeatedly to the expectations from the vendors that FPGAs will "eat away at the ASIC and ASSP markets." I don't doubt that this is true around the edges, but when vendors say they are no longer fighting for their share of a $5 billion PLD market and instead are attacking a $58 billion market, that's when my cynicism takes over.
The ASIC market has unquestionably been eroded by FPGAs to the point where a large percentage of the remaining 3,000 to 4,000 design starts each year are either bleeding-edge massive chips or mixed-signal designs using less aggressive (and cheaper) process nodes. Consider the biggest-selling consumer gadgets on the planet: smartphones, laptops, and tablets. There are not many FPGA sockets in those. I'd concede that sometimes TVs and digital cameras include an FPGA to support a specialist interface, but these sockets are usually transitory until the next ASSP spin sucks in the new design. Sorry, but I just don't see ASSP vendors such as Qualcomm and Broadcom looking over their shoulders at Xilinx and Altera and quaking in their boots.
What is new on the Programmable Planet today is the advent of so-called 2.5D IC products, which are also known as active-on-passive 3D ICs. These are combinations of dice mounted on interconnecting substrates. (See: The State of the Art in 3D IC Technologies.)
Xilinx stole a march on Altera (and lots of headlines) with the 2000 LUT device from the Virtex-7 family, which is built up from four dice using a stacked silicon interconnect (SSI) technique. This device, which is now shipping, offers 2 million logic cells, or twice the capacity of the largest Altera offering. This 2.5D technology opens up the possibility of mixing different types of dice, such as memory or analog chips alongside FPGAs. The mixed technology products have endless potential, but the cost will not be low.
But all of this is today. What of the future?
Next page >