FPGA Design Services: What a Project Involves and How to Choose a Partner

Written by Sciotex Engineering
In this article

FPGA design services cover the engineering needed to turn a timing-critical requirement into working logic on a field-programmable gate array: architecture, HDL or graphical design, simulation, timing closure, hardware bring-up, host software and test. Companies hire an FPGA design partner when a job needs deterministic timing, massive parallel I/O or data rates that a processor alone cannot deliver reliably.

If you are new to the technology, start with our overview of what an FPGA is and where it is used, and our comparison of GPU vs CPU vs FPGA for imaging and AI. This article is for engineers and managers who already know they may need FPGA development services and want to know what a project involves, what to expect from it and how to choose a partner.

When a project actually needs an FPGA

An FPGA is not the default answer. It costs more engineering time than software on a processor, so it should be chosen for a reason the table makes clear.

RequirementCPU or microcontrollerGPUFPGA
Deterministic response in microseconds or nanosecondsHard on a general-purpose OS; possible on a real-time controller for slower loopsNo, built for throughput, not latencyYes, timing is fixed by the clock and the logic
Many parallel I/O channels sampled at onceLimited by interrupts and busesNot designed for direct I/OYes, every channel gets its own logic
Custom or legacy interfaces and protocolsOnly if hardware already supports themNoYes, the interface is implemented in logic
Continuous high-rate data streams, such as cameras or fast ADCsCan drop data under loadGood once data reaches itYes, and it can reduce data before it reaches the host
Large floating-point and AI workloadsSlowBest choicePossible, but usually more effort than a GPU
Development time and costLowestModerateHighest; needs specialist skills and verification

A common good design splits the work: the FPGA handles triggering, timing and data reduction, and a PC or real-time controller runs the user interface, storage and higher-level logic.

The FPGA design flow, step by step

  1. Requirements and interface specification. Clock rates, latencies, channel counts, data formats and every external interface, written down with numbers.
  2. Architecture and resource estimate. Clock domains, pipelines, memory buffers and an estimate of logic, DSP and block RAM use, which drives device selection.
  3. Design entry. VHDL, Verilog or SystemVerilog for most logic, high-level synthesis for some algorithms, or LabVIEW FPGA where the system is built on NI hardware.
  4. Simulation and verification. Testbenches that exercise each block and the whole design, including error and corner cases, before anything is loaded into hardware.
  5. Synthesis, place and route, and timing closure. The tools map the design onto the device; the engineer fixes any paths that miss timing.
  6. Hardware bring-up and integration. Loading the design onto the board, checking signals with a scope or logic analyzer, and connecting the host through drivers and DMA.
  7. System test, documentation and handover. Proving the full system meets the specification and delivering everything needed to maintain it.

Steps 2 and 4 decide most outcomes. A design that is more than about 80% full is much harder to close timing on, and bugs found in simulation take minutes to fix, while bugs found on the bench can take days.

Worked example: why the FPGA should reduce the data

Worked example: an 8-channel, 10 MS/s acquisition system

Eight channels, each sampled at 10 million samples per second with 16-bit (2-byte) samples: 8 × 10,000,000 × 2 bytes = 160 MB per second of raw data, every second the system runs.

If the application only needs an RMS value for each channel over every 1,000 samples, the FPGA can compute it in real time. Output = 8 channels × (10,000,000 ÷ 1,000) = 80,000 values per second; at 4 bytes each, that is 320 KB per second.

That is a 500× reduction (160 MB ÷ 0.32 MB). The host no longer needs a high-bandwidth bus or fast storage just to keep up, and no samples are lost while the PC is busy.

At a 100 MHz FPGA clock each cycle is 10 ns, so a trigger or control response of 1 µs is a fixed 100 clock cycles, repeatable every time.

This pattern, processing at the edge and sending results rather than raw samples, is central to high-speed data acquisition and to precise instrument timing such as synchronizing lasers, galvos and cameras in research microscopes.

What deliverables to expect

Agree on deliverables before the project starts. You should be able to rebuild, modify and test the design without the original developer.

DeliverableWhy it matters
Complete source code and build scriptsYou can rebuild the bitstream on your own tools and version-control it
Constraint files (pins, clocks, timing)Without them the same source will not produce a working design
Testbenches and simulation resultsProof of what was verified, and the basis for testing future changes
Timing and resource reportsShow how much margin is left for future features
Interface control documentDefines registers, data formats and protocols for the host and other teams
Host driver or API and example codeLets your application software use the FPGA without reverse engineering it
Test procedure and results on hardwareShows the delivered system met the specification
Clear IP ownership termsConfirms you own the custom design and lists any licensed third-party cores

What drives cost and schedule

  • Number and complexity of interfaces, especially high-speed serial links and custom protocols.
  • Clock domains and timing margins: more domains and tighter timing mean more verification and closure work.
  • Off-the-shelf vs custom hardware: a commercial FPGA board avoids PCB design, layout and board bring-up; a custom board adds them.
  • Depth of verification required by the application, and any industry standard the design must follow.
  • Host software and system integration: drivers, user interface, data storage and links to other equipment.
  • How well the requirements are defined: changing requirements late is the most common cause of FPGA schedule overruns.

How to choose an FPGA design partner

Many firms can write HDL. Fewer can deliver a working system. These questions separate the two:

AskWhy it matters
Which FPGA families and tool chains have you shipped designs on?Experience on your target device shortens bring-up
Do you deliver complete systems, or logic only?Most problems appear where the FPGA meets the board, sensors and host software
How do you verify designs before hardware?Weak simulation shows up later as long bench debugging
Can you show examples similar to ours, even if anonymized?FPGA work is often under NDA, but a partner should be able to describe comparable projects
Who owns the design, and what do we receive at handover?Avoids being locked to one developer
Who supports it after delivery, and how quickly?FPGA systems often run for many years and need updates

Sciotex has built many custom FPGA-based systems for test, data acquisition, imaging and control, including high-speed laser triggering and galvo mirror control for microscopes and a ring laser gyro test system. As an NI and LabVIEW integrator, we develop in LabVIEW FPGA as well as HDL, and we build the host software, instruments and mechanics around the logic so the whole system works together. Much of this work is confidential, so we are glad to discuss comparable projects directly.

Need an FPGA development partner?

Sciotex provides FPGA programming services and complete FPGA-based systems, from specification and logic design to host software, test and long-term support, designed, programmed, built and supported in the US since 1997.

Schedule a free FPGA consultation

Frequently asked questions

What are FPGA design services?

FPGA design services are engineering services that take a requirement through FPGA architecture, HDL or graphical design, simulation, timing closure, hardware bring-up, host software and test, to deliver working FPGA logic or a complete FPGA-based system.

When should I use an FPGA instead of a processor?

Use an FPGA when you need deterministic timing in microseconds or nanoseconds, many I/O channels handled in parallel, custom interfaces, or real-time processing of high-rate data streams. For general logic and user interfaces, a processor is cheaper to develop.

What languages are used for FPGA development?

Most FPGA logic is written in VHDL, Verilog or SystemVerilog. High-level synthesis from C or C++ is used for some algorithms, and LabVIEW FPGA is used on NI hardware.

What should I receive at the end of an FPGA project?

Source code and build scripts, constraint files, testbenches and simulation results, timing reports, an interface control document, host drivers or an API, hardware test results and clear IP ownership terms.

How long does an FPGA design project take?

It depends on interfaces, verification depth and whether custom hardware is needed. Logic for an existing board can take weeks; a complete system with custom hardware can take months. Clear requirements shorten every phase.

Related articles

Written by

Sciotex Engineering

The Sciotex engineering team has designed and built industrial, test and scientific software since 1997.

Have a software project or legacy system?

Talk it through with a Sciotex engineer.

In this article
Need help with industrial software?

Talk it through with a Sciotex engineer.

Next step

See what modern, supportable software could do for your system