field-programmable gate array

A field-programmable gate array is a chip containing thousands of unconfigured logic blocks and interconnects that you wire together after manufacture by loading a bitstream into its memory. Unlike application-specific integrated circuits (ASICs), which are hard-wired at the foundry, an FPGA lets you define which blocks talk to which without touching hardware. This matters enormously when you need to deploy logic that will change, debug, or adapt to new specifications without ordering a new run of custom silicon.

The logic blocks themselves are typically look-up tables (LUTs), usually 4-input or 6-input, combined with flip-flops and multiplexers. These sit in a grid pattern, surrounded by programmable routing channels. You configure everything using hardware description languages like VHDL or Verilog. Synthesis tools compile your code into a bitstream, a binary file that sets every switch in the array. The bitstream loads into static RAM cells, or sometimes flash memory, that control the routing and function of each block.

Where FPGAs fit in production and prototyping

In industrial settings, FPGAs appear in motion control, real-time signal processing, and custom I/O protocols where latency or bandwidth matter too much for software on a CPU. A typical industrial FPGA might have a few thousand LUTs and run at 100 to 400 megahertz. Larger devices in the same family scale to hundreds of thousands of LUTs. You trade off cost per unit (FPGA chips cost more than equivalent ASICs at volume) for flexibility and fast time-to-market. Reconfiguration happens either at power-up or, in advanced devices, at runtime while the rest of the system runs.

The main failure modes are configuration memory corruption from radiation or power faults, logic timing violations that arise when reconfigured designs exceed propagation delays, and thermal runaway if power dissipation during reconfiguration is not managed. Industrial designs often include watchdog circuits and periodic bitstream reloads to catch memory bit flips. Vendors like Xilinx and Intel Altera dominate the market; smaller players serve niche applications.

The term 'field-programmable' emphasizes that you configure the part in place, after it has left the factory and entered the field, rather than before shipment. This contrasts with mask-programmable gate arrays, which required factory intervention to set metal masks, and with ROM-based or EPROM-based logic, which are slower and less flexible. The FPGA invented in 1985 eliminated that constraint and became the standard flexible silicon platform for custom logic that does not justify an ASIC run.