FPGA Programming with Linux
Free software licenses and operating systems like GNU/Linux make it possible to learn programming and customize state-of-the-art software in countless ways. Hacking software, however, isn't the last frontier anymore. What if you could hack integrated circuits directly—that is, tell a chip to connect its internal transistors to create exactly the custom, real-time digital hardware you want? This is precisely what you can do with Field Programmable Gate Arrays (FPGAs). In this article, I explain how to do it with nothing else but your Linux computer and an inexpensive development board.
This isn't the first time Linux Journal has covered FPGAs (see Resources), but these devices have made huge progress since those articles were written. Plus, prices for powerful development boards have come down dramatically. Today, you can do really cool things with a $200–$300 FPGA-based board, a typical personal computer and one square foot of desk space. Finally, the FPGA community is now big enough and stable enough to make life much easier for beginners. The major FPGA makers offer text or video tutorials and forums where even newbies can find support. Above all, Web sites like Opencores.org codevelop and release Linux-compatible FPGAs and boards like the EUS 100LX with the same spirit (and licenses) as free software. In summary, the barrier to entry is much lower now than it was even just a couple years ago, making FPGA design both a cool hobby and an affordable and interesting addition to the offerings of even high schools.
A digital integrated circuit (IC) is a chip that deals only with binary digits—meaning signals that can assume only one of two states: 0 or 1, high or low voltage and so on. An FPGA is an IC consisting of one array of digital logic gates. These basic circuits, made up of a few transistors each, instantiate either a flip-flop or lookup tables capable of implementing any boolean function of up to four binary signals. The magic of FPGAs is that the connections among the logic gates (the actual circuit you need) are made at power-up by reading the configuration instructions written into a bit file. Changing the file changes the function of the FPGA.
As flexible as they are, microprocessors are always and only microprocessors: single-purpose hardware that is capable only of executing (relatively slowly) instructions in one machine language. An FPGA, instead, becomes whatever hardware you need. It can morph into a microprocessor, a game console, a real-time IP switch or encryption device, an antitheft server or anything else you can imagine. The only limits are that your circuit cannot require more transistors or external pins than those physically present on the chip, and it can't go faster than the intrinsic propagation delay from gate to gate.
That said, modern FPGAs are powerful enough to let you squeeze several Linux-compatible microprocessors, like the Nios, the PowerPC or the Microblaze, inside them and still have lots of room for your own custom circuits. In many cases, you can load certified CPU designs from libraries and place them on silicon with simple commands, creating very flexible, complete systems inside just one chip.
The major manufacturers of FPGAs and other programmable ICs are Altera and Xilinx, followed by Lattice and Atmel. Although the example in this article uses Xilinx products, the general procedure is the same with all vendors, and all of them have similar boards. In all cases, the design software is closed-source and often expensive, but it is possible to download either free trial versions valid for one or two months, or free versions with reduced functionality but free upgrades and no expiration date.
The current way to design FPGAs is to write a behavioral model in a Hardware Description Language (HDL), like Verilog or VHDL, which supports concurrency and synchronous circuits. Concurrency allows you to create fully parallel, independent processes, each describing how to update some variables continuously. Synchronous circuits, instead, are those made of flip-flops that change their state only on the edge of some clock signal.
After the design has been written and verified with an HDL simulator, a compiler creates a list of all the logic gates and the wires (nets) that must connect them to reproduce the functionality of the HDL model. After this logic synthesis, layout programs read the netlist and several constraints files to find out which logic gates inside the FPGAs must be used and which physical, internal wires must connect them to each other. The end result is the bit file that the FPGA reads at power-up.
The official Xilinx design suite is called ISE Foundation (www.xilinx.com/ise), and the reduced functionality version is called Webpack. Both programs run on Windows, Red Hat Enterprise and SUSE Linux Enterprise (32- or 64-bit). Other Linux distributions may work too, but there is no guarantee.
ISE has a graphical installer where you must accept the software license and enter the key you got after a free registration on the Xilinx Web site. After it's finished, you'll find a script called settings32.sh or settings64.sh in the installation directory—that's the one you have to source to add the Xilinx software to your path. After this, type ise at the prompt to launch the Project Navigator (Figure 1). This is a front end to a bunch of specialized programs, one for each design phase. You also can run most of these back-end utilities from the command line. The Navigator includes a Tcl prompt and, if you select Project→Generate Tcl Script, it will save all the commands you entered through the GUI as a Tcl script.
Other ISE components, like the simulator, the FPGA Editor and ChipScope, have graphical interfaces. You'd use the FPGA Editor to place and connect single gates manually when the software fails to do it according to your specs. ChipScope is like a software oscilloscope with a USB probe. During synthesis, you can add special circuits to your design that will buffer the internal signals you want to see and send them over a USB cable to the ChipScope software for display. We'll see the ISE HDL simulator at work in a moment.
The board I got from Xilinx for this article is the Spartan-3AN Starter Kit (Figure 2), based on the Spartan XC3S700AN FPGA (Figure 3), which contains about 700K system gates. Around it there are several memory chips, a 50MHz onboard clock, a connector for an external clock and several extra components, from D/A and A/D converters to generic I/O pins, assorted LEDs, sliders and push buttons and, finally, a two-line LCD display. The ports (Figure 4), are enough to make a full-custom Linux PC out of this board: 10/100 Ethernet/PHY, USB, keyboard, VGA, serial and stereo mini-jack for PWM audio. A universal power adapter and USB cable are included, as are four different bit files that demonstrate the capabilities of the FPGA. The corresponding design files are freely downloadable from the Xilinx Web site.
Articles about Digital Rights and more at http://stop.zona-m.net CV, talks and bio at http://mfioretti.com