Altera, OQC and Nvidia build an FPGA controller design linking a quantum processor to GPUs

AI & Compute

Altera, OQC and Nvidia Link Quantum Processor to GPUs via FPGA

Altera, OQC and Nvidia showed an FPGA-based quantum controller at SC26, pairing an Agilex 9 Direct RF chip with GPUs over NVQLink. Public release is set for January 2027, with no latency figures disclosed.

By
Sophie Lindqvist
Filed
Channel
AI & Compute
Read
3 min read

Altera says it has built an end-to-end quantum system controller reference design with Oxford Quantum Circuits and Nvidia, running real-time qubit control on an Agilex 9 Direct RF FPGA and connecting it to Nvidia GPUs over the NVQLink interconnect. The companies announced the work on Nov. 16. Altera expects to release the reference design in January 2027.

The three firms are demonstrating the full system this week at Altera's booth at SC26, the supercomputing conference in Chicago. A booth demo is not a shipped product. The January 2027 release date is a company expectation, not a confirmed availability.

What does the controller actually do?

Superconducting qubits are steered and read with precisely timed microwave pulses. The classical electronics that generate and measure those pulses usually fill racks of separate radio-frequency chains, converters and cabling. Altera said qubit counts are growing faster than the space available for that equipment, so control hardware has to get denser.

The Agilex 9 Direct RF chips have analog-to-digital and digital-to-analog converters built in. That integration means control and readout signals need no external up- or down-conversion. Frequency multiplexing lets several readout channels share one RF port.

In the reference design, the FPGA handles the time-critical work:

  • qubit control,
  • resonator readout,
  • spectroscopy, with synchronized scheduling.

Heavier processing runs on the GPUs. That includes calibration and machine-learning-based classification of readout signals. CUDA functions drive the FPGA logic, according to the announcement.

The FPGA drives a new-generation OQC quantum processing unit through radio-frequency conditioning electronics built by OQC. "Precise control and fast feedback are essential to unlocking the capabilities of our quantum processors," said Simon Phillips, OQC's chief product officer.

Sam Stanwyck, Nvidia's director of quantum product, said NVQLink gives quantum builders "an open architecture" for AI-driven calibration and readout. Nvidia's role is the interconnect and the GPU-side software. Altera's role is the real-time control silicon. OQC supplies the quantum processing unit and the RF conditioning electronics between it and the FPGA.

What do developers get?

The reference design ships with a validated set of functions and a software stack, the companies said. The stack covers:

  • qubit and resonator spectroscopy,
  • qubit initialization,
  • readout,
  • a GPU-based machine-learning readout classifier,
  • an emulation mode for testing before qubit hardware is available.

The emulation mode matters for teams without their own quantum hardware. It lets them exercise the control pipeline first. Whether the validated functions cover enough of real calibration workflows to shorten deployment is something the January 2027 release will have to show.

What is not in the announcement?

The companies did not publish latency figures. That omission is notable for a design whose stated purpose is precise control and fast feedback. There is also no qubit-count target, no pricing, and no named customer beyond the three partners themselves.

The announcement describes a new-generation OQC quantum processing unit but gives no specification for it. No throughput or readout-fidelity numbers accompany the machine-learning classifier. Each of these gaps leaves the performance claims untestable until independent users get hardware and software in hand.

Who is Altera?

Altera is the FPGA maker majority-owned by Silver Lake. The company was formerly Intel's FPGA unit before the private-equity firm took a controlling stake. The quantum controller design is a reference design, not a product line, and its commercial terms remain undisclosed.

Why does the architecture split matter?

The design draws a clear line between two classes of work. The FPGA does the deterministic, timing-critical signal generation and capture. The GPUs do the compute-heavy calibration and the classification of noisy readout signals, connected through NVQLink. This division lets each processor type do what it is structurally good at, and it ties quantum control into Nvidia's broader accelerated-computing stack.

If the January 2027 release delivers the promised functions and the emulation mode as described, quantum labs will gain a denser, GPU-coupled control path that replaces racks of discrete RF chains; until latency and fidelity numbers appear, the density argument is the only measurable claim on the table.

Source: mlq.ai

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,052 articles

Related articles

« Previous articleNext article »