Building Whole DSP Systems in Software with xcore®.ai

In today’s digital world, devices from smart speakers to wearable health tech demand reliable, flexible, high-performance digital signal processing (DSP). But designing full DSP systems often involves juggling specialised hardware, complex integrations, trade-offs between latency, power and cost — and long development cycles.

Enter xcore.ai , XMOS’s answer to these challenges. With it, you can now build entire DSP systems in software alone, bringing together DSP workloads, I/O, control logic — even AI — without the complexity of stitching together multiple discrete chips.

Here are the key pillars that make xcore.ai stand out:

Flexibility

Whether you’re processing in fixed-point or floating point, you’ll get the flexibility to integrate I/O protocols, control algorithms, AI models — all within a single software-defined platform. That reduces the bill of materials and simplifies the hardware design.

Predictable Programmability

The architecture is multi-threaded, multicore, with deterministic behaviour. For applications that require hard real-time performance — multiple sample rates, tight timing, low latency — this matters. xcore.ai ensures that different parts of the system don’t interfere with each other.

Feature-Rich Capabilities

Each xcore.ai device has:
– 16 hardware threads (HART) across two processor tiles
– 512 kB SRAM per tile
– An integer vector unit capable of efficient block floating point operations
All of this resides inside a single homogeneous development environment.

Here’s what xcore.ai delivers:

MetricValue
Sustained FLOPS~800 MFLOPS
Peak FLOPS~1,600 MFLOPS
FFT throughput~98,561 FFTs/sec (fixed point), ~11,300 float FFTs/sec
FIR filter performance~957 million taps/sec
IIR filters (per biquad)~251 million/sec

If your DSP needs scale beyond what one chip can handle, XMOS also supports multi-chip systems via dedicated links.

Beyond the raw specs, what often slows projects down is building out the standard DSP building blocks from scratch. XMOS alleviates this by offering:

  • A library of fundamental DSP building blocks
  • High-level reference applications: PDM interfaces, acoustic echo cancellation, noise suppression, asynchronous sample-rate conversion, automatic gain control, and more
  • Modifiable, extendable codebases so you can tailor to your specific system or interface needs — all within one cost-efficient, embedded solution.

This platform is perfect for many use cases including:

  • Personal electronics & smart home devices
  • Computer peripherals
  • Gaming & immersive audio
  • Home theatre / audio systems
  • Communications gear
  • Wearables
  • Industrial & automotive sectors (pro-audio, motor control, infotainment…)

Whatever the domain, if your product needs real-time DSP, low latency, flexibility, and cost-efficient hardware, xcore.ai is worth a close look.

Whether you’re evaluating DSP architectures or starting your next embedded project:

Consider how single-chip vs multi-chip scaling matches your performance needs and cost constraints.

Explore XMOS’s Application Notes to see what you can adapt.

Take a look at the XMOS developer tools — ramp-up is smoother when you can build from working templates.

Scroll to Top
Secret Link