FPGA vs MCU: When to Use Which in Embedded Design
FPGA vs MCU: When to Use Which in Embedded Design
Every embedded engineer hits this fork in the road sooner or later: should this design use a microcontroller or an FPGA? The two devices might share a PCB footprint and a power rail, but under the hood they operate on fundamentally different principles. Choosing wrong means wasted BOM cost, bloated firmware, or a board that simply cannot meet timing. Here is a practical framework for deciding which architecture fits your project.
Architecture: Sequential vs Parallel
A microcontroller (MCU) executes instructions one after another. An ARM Cortex-M4 at 180 MHz is fast, but it still processes your C code line by line. An FPGA, by contrast, is a sea of programmable logic blocks that all operate simultaneously the moment power is applied. That hardware parallelism is the single biggest differentiator: ten independent signal-processing chains on an FPGA run in true parallel, while an MCU must time-slice them through an RTOS scheduler.
This architectural gap defines the use-case boundary. If your system reads a few sensors, runs a control loop, and drives a display, an MCU is the natural fit. If you need to ingest 16 channels of ADC data at 100 MSPS and apply real-time FIR filters to each, only an FPGA can keep up.
Head-to-Head Comparison
| Parameter | Microcontroller (MCU) | FPGA |
|---|---|---|
| Processing Model | Sequential, instruction-driven | Parallel, hardware-defined |
| Typical Clock Speed | 48–480 MHz | 100–500 MHz (fabric) |
| Latency | Microseconds to milliseconds | Single clock cycle (nanoseconds) |
| Development Language | C/C++, Rust, MicroPython | VHDL, Verilog, HLS (C++ to RTL) |
| Unit Cost (entry-level) | $0.50–$15 | $5–$200+ |
| Power Consumption | Low (mW to 1 W) | Moderate to high (1–15 W) |
| Reconfigurability | Firmware update only | Full hardware reconfiguration in field |
When an MCU Makes More Sense
Cost-sensitive consumer products. When the BOM target is under $10, an STM32F103 or GD32F103 at $1.20 is hard to beat. Even the cheapest FPGA (Gowin GW1N-1) starts around $4 in volume.
Protocol-heavy designs. MCU peripherals — UART, I2C, SPI, CAN, USB — are hard IP blocks that work out of the box. Implementing a full USB stack in FPGA fabric is weeks of engineering time.
Battery-powered devices. Modern MCUs draw single-digit microamps in deep sleep. An FPGA needs power-on sequencing and always burns static leakage current, making sub-mW operation challenging.
Small firmware teams. Writing RTL requires a different skill set than embedded C. If your team has no HDL experience, the learning curve alone can derail a schedule.
When an FPGA Is the Right Call
High-speed parallel I/O. Anything involving multiple MIPI camera streams, LVDS ADC interfaces, or multi-gigabit transceivers is FPGA territory. An MCU simply lacks the pin bandwidth.
Deterministic ultra-low latency. Motor control loops that must react within 100 ns cannot tolerate OS scheduler jitter. Hardware state machines in an FPGA respond in one clock cycle, every cycle.
Future-proof field upgrades. If a protocol standard changes (think 5G NR evolving to 6G), an FPGA can be reprogrammed to support new waveforms. An ASIC or fixed-function MCU requires a board respin.
Signal processing pipelines. Digital down-conversion, FFT, and custom filtering run orders of magnitude faster in parallel hardware than in sequential software. A Xilinx Artix-7 can process 32 channels of DDC simultaneously at 200 MSPS.
The Hybrid Middle Ground
Modern SoC FPGAs — Xilinx Zynq-7000, Altera Cyclone V SoC, and Microchip PolarFire SoC — integrate an ARM Cortex-A or RISC-V processor alongside FPGA fabric on a single die. These devices let you run Linux on the processor for networking and UI while offloading DSP or custom I/O to the programmable logic. For applications like software-defined radio, medical imaging, or industrial vision, a hybrid SoC FPGA often delivers the best of both worlds at a mid-range price point.
Choosing between FPGA and MCU is rarely about which technology is superior. It comes down to whether your design bottleneck is cost per unit, development velocity, power budget, or raw processing bandwidth. Map your requirements against these four axes, and the right architecture usually announces itself.
www.aplusic.com
