Findings: FPGAs Favored over ASICS

Hi Chris

I have been using Xilinx FPGAs ( http://www.xilinx.com )since 1987, all families 2000, 3000, 4000, and Spartan families, they can be reprogrammed on the fly.

What I like most using FPGAs is that math processes can go in parallel, and many signal processes run at once, for example if you need to calculate fast a high resolution polynomial equation to linearize a two variable sensor, something like

X(t) * ( X(t) * ( X(t) * ( X(t) * Y4(t) + Y3(t) ) + Y2(t) ) + Y1(t) ) + Y0(t)

where X(t) and all Yn(t) are high resolution signals and parameters sampled at fast rate

then, instead of processing a serial sequence of instructions like in any Von Newman CPU, in a Xilinx FPGA you use 4 floating point multipliers and 4 floating point adders, and for each signal sample you calculate all products and additions within the period between two clock ticks, and you use a single latch to freeze only the final result, you don't worry about inner results, just the last, and even more, you can get the result sooner using an asynchronous flag that indicates calculations are ready. The READY flag jumps when no more transients appear, and the trick is to detect every transient with wide XOR gates monitoring all bits of the last result, and using every XOR glitch to trigger a retrigerable "One Shot"; when the output of this "One Shot" gets finally LOW, then you can be sure all the real process is finished, and that may be much before the next clock tick, so using some statistics you will know safely which is the real maximum clock frequency and signal sampling rate to drive the Analog to Digital converters for X(t) and Y(t).

I prefer graphic means to design logic, I mean using graphic symbols, and functional blocks nested in a hierarchical structure, where at any level all symbols are at the same hierarchical level. Some designers use textual languages like VHDL, but I think that VHDL does not allow me to have a panoramic picture of the process. I prefer graphics.

Another thing is that you can invent your own mathematics, for example instead of using a mantissa and an exponent to represent IEEE 754 floating point math, I use a single number to represent real numbers within an almost infinite dynamic range, say +/-1E-300 to +/-1E+300. Years ago I tried to present this idea to the US PTO, but at that time I was very poor and didn't have money to pay for the revisions, and later I presented this idea to Xilinx Inc, but they didn't express any interest, so finally I take the decision of just to use this idea in all my projects. Also you can invent new operations, you already have X*Y, X+Y, X-Y and X/Y, but also in floating point complex numbers, the "operation" X//Y = X*Y/(X+Y) is very important.

Jaime Soto Figueroa
http://www.matharts.cl/

In the digital world FPGA's will rule, but in the analog world, especially at automotive voltages (up to 36 volts with transients), the ASIC is still the only option. There are some semi-custom options that that facilitate low cost and speedy development, but programmability after installation is not available. I have seen some mixed signal FPAA, but the voltage capabilities are still at 5 volts.