FPGAs and Hardware Acceleration

A normal trading program runs as software: instructions stored in memory, fetched and executed one at a time by a general-purpose CPU, with an operating system, network stack, and application code all adding delay before a decision reaches the exchange. A field-programmable gate array (FPGA) is a chip whose internal circuitry can be reconfigured to directly implement a specific piece of logic — instead of a CPU interpreting instructions to decide "if price crosses this level, send an order," an FPGA can have that exact decision wired into hardware, executing in nanoseconds rather than microseconds.

Firms competing on raw speed (particularly market makers and latency arbitrageurs) use FPGAs to handle the most time-critical steps — parsing incoming market data and firing an order — while leaving less time-sensitive logic like strategy research or risk limits in ordinary software. The tradeoff is development cost and inflexibility: writing and testing FPGA logic requires specialized engineering skill distinct from normal software development, and changing the logic means reprogramming the chip rather than editing and redeploying code.

Worked example. A software trading path might take 5–20 microseconds to react to an incoming quote update once network delay is stripped out; an FPGA implementing the same reaction directly in hardware can bring that down to under 1 microsecond, a difference that matters when competing strategies are racing for the same fleeting price discrepancy.