Q. What are the advantages of integrating memory on the chip and not as an external element as used to be the case?
A. Typically, if you look at a board, you have a chip, a processor, FPGA etc and then, you have the memory, either a D-RAM or an S-RAM. However, the biggest bottleneck is the bandwidth between memory and chip; how much data you can buffer, process and take out. Now, that is totally on the limitation of the board. Now imagine us taking the memory inside the chip. It’s a real high capacity memory which is now inside a ship which translates to higher bandwidth. That’s the most significant use of integrating the memory on the chip which is what we’ve done on the 20nm. Before the 2nd generation, we never had memory on the chip. Now, by adding memory, we have ensured that the designers don’t have to focus on the memory aspect at all. On the transceivers part, earlier we used to do 28 Gbps, now we’re going from 33 to 56 Gbps transceivers.
Q. So, what are the performance improvements Xilinx is looking at from this second generation portfolio?
A. From an FPGA perspective, at 20nm, we are going to see a 30-50% improvement on the price per performance per watt front. What that means is 30-50% price reduction, performance improvement and power reduction, all for the same capability. Our analog mixed signal will become mainstream. Right now, we’ve had a very significant capacility at 28nm and in addition a lot of IP subsystems will be integrated into it.
From an SoC standpoint, we only had dual-core cotex A9. Now, we are introducing heterogenous multi-core; that means there is not just the ARM core inside but also a GPU core. Plus, the FPGA processing and a software environment which suits it.
For 3D-ICs, it will have the FPGA die, I talked about the tranceivers that we’ve integrated, now we are also integrating wide memory onto the 3D-IC.
Q. How does Xilinx differentiate itself from its competitors in the FPGA market? What would you say sets Xilinx apart?
A. If you compare an FPGA company like us and companies who make standard products, the basic benefit we provide to the customers is programibility and flexibility. So, when we design our product portfolio, we make sure that we are giving them the right amount of resources at the right pricepoint with the power, performance benefits included. For every node we have, we come up with a portfolio of devices at the right gate density and we look at different applications like for example if it’s a 100 Gbps network block or a flat-panel TV, what kind of capabilities would they need. So, we need to decide on a wide array of applications bu really optimise a set of devices which kind of covers most of the applications. A standard product provider would make a chip for a TV and say that this chip is used for image processing for a TV and nothing else. But our FPGA will do that and then it can also be used for military applications. The beauty of FPGAs is that, the same product is used across different vertical markets. Thus, we have a variety of different customers for the same product.