Fusionality: Standardized Controls for Fusion Reactors

The Fusion Control Gap and Fusionality’s Mission

Magnetic‑confinement fusion—whether via tokamaks, stellarators, or compact spherical devices—relies on ultra‑precise regulation of plasma temperature, shape, and fuel density. While the physics of confinement is a research frontier, the engineering of the control loop is surprisingly uniform across projects. Industry insiders estimate that about 80 % of a fusion reactor’s control system is functionally identical from one company to the next.

Enter Fusionality, a Lausanne‑based startup founded by former Google DeepMind researchers Federico Felici (CEO) and Jonas Buchli (CTO). Their vision is simple yet disruptive: provide a plug‑and‑play suite of hardware and software that covers the common 80 % while allowing each fusion venture to focus its engineering effort on the remaining 20 % that differentiates its design. By “speaking the language of fusion,” Fusionality hopes to eliminate the costly, time‑consuming custom‑build approach that currently dominates the sector.

Why Standardized Controls Matter for the Fusion Supply Chain

Reducing Capital Expenditure

Building a bespoke control system from the ground up can cost tens of millions of dollars in engineering labor, component procurement, and validation testing. For early‑stage fusion startups, this expense competes directly with plasma‑physics research budgets. A standardized platform lowers the entry barrier, enabling more players to allocate capital toward core scientific challenges rather than peripheral hardware integration.

Accelerating Time‑to‑Experiment

Every iteration of a control loop requires extensive simulation, hardware‑in‑the‑loop testing, and safety certification. Fusionality’s pre‑validated modules cut the iteration cycle from months to weeks, allowing companies like Commonwealth Fusion Systems or Realta Fusion to run longer plasma shots and gather data faster.

Enabling a Modular Ecosystem

The “Lego‑block” philosophy—starting with a tightly selected core set of technologies and expanding the library over time—creates a marketplace where third‑party vendors can contribute add‑ons (e.g., advanced diagnostics, specialized power electronics). This modularity mirrors successful ecosystems in the semiconductor and aerospace industries, where standard interfaces drive rapid innovation.

Technical Breakdown of Fusionality’s Control Systems Suite

Core Hardware Stack

• Real‑time FPGA Controllers – Low‑latency processing units capable of sub‑microsecond response times, essential for magnetic field adjustments.
• Modular Power‑Conversion Units – Scalable converters that handle the high‑current demands of superconducting coils while providing precise voltage regulation.
• Sensor Interface Boards – Unified drivers for Langmuir probes, magnetic pick‑up coils, and infrared cameras, exposing a common API to higher‑level software.

Software Layer and Simulation Environments

1. Control Logic Framework – A C++/Python hybrid that abstracts hardware specifics behind a deterministic state‑machine model. Developers can plug in custom algorithms for plasma shaping without rewriting low‑level drivers.
2. AI‑Enhanced Optimization Modules – Rather than handing over full control to an AI, Fusionality integrates machine‑learning models that suggest optimal set‑points for coil currents, anticipate disruptions, and fine‑tune fueling schedules. This approach respects safety constraints while leveraging DeepMind‑style reinforcement learning techniques.
3. Digital Twin Simulators – High‑fidelity, GPU‑accelerated models that replicate plasma dynamics and hardware response. Engineers can run “what‑if” scenarios offline, dramatically reducing the need for costly physical trial runs.

Integration Workflow

1. Select Core Modules – Choose the appropriate FPGA board, power unit, and sensor package based on reactor size.
2. Configure the Control Logic – Use the provided SDK to map reactor-specific actuators to the generic control API.
3. Run the Digital Twin – Validate the configuration against simulated plasma behavior.
4. Deploy to Hardware – Upload the compiled control code to the FPGA, connect power modules, and begin live operation.

Market Landscape: Target Customers and Competitive Position

Fusionality’s immediate addressable market consists of magnetic‑confinement startups that have already secured funding but lack in‑house control expertise. Notable prospects include:

• Commonwealth Fusion Systems – Working on a high‑temperature superconducting tokamak.
• Realta Fusion – Developing a compact, spherical tokamak for commercial power.
• Proxima Fusion – Focused on high‑beta plasma configurations.
• Type One Energy – Pursuing a low‑cost, modular fusion platform.

These companies share a common pain point: the scarcity of vendors that understand both the **physics of plasma

physics of plasma and the demanding real‑time control hardware that keeps it stable. Most existing suppliers specialize in either high‑energy physics instrumentation or generic industrial automation, leaving a gap for a provider that can bridge both worlds. Fusionality positions itself precisely in that niche, offering a “fusion‑first” stack that speaks the language of tokamaks, stellarators and emerging compact concepts alike.

Competitive Edge

🔹 ---------
• Fusionality: -------------
• Traditional Industrial Automation: -----------------------------------
• Pure‑Play Fusion Labs: -----------------------

🔹 *Fusion‑specific sensor drivers*
• Fusionality: ✔︎
• Traditional Industrial Automation: ✘ (generic I/O)
• Pure‑Play Fusion Labs: ✔︎ (custom, but limited)

🔹 *Real‑time FPGA with sub‑µs latency*
• Fusionality: ✔︎ (pre‑tuned for coil currents)
• Traditional Industrial Automation: ✔︎ (but not plasma‑optimized)
• Pure‑Play Fusion Labs: ✘ (built from scratch)

🔹 *AI‑assisted set‑point recommendation*
• Fusionality: ✔︎ (reinforcement‑learning trained on tokamak data)
• Traditional Industrial Automation: ✘
• Pure‑Play Fusion Labs: ✔︎ (research prototypes)

🔹 *Modular, plug‑and‑play hardware*
• Fusionality: ✔︎ (standardized connectors, API)
• Traditional Industrial Automation: ✔︎ (but not fusion‑ready)
• Pure‑Play Fusion Labs: ✘

🔹 *Digital twin simulation suite*
• Fusionality: ✔︎ (GPU‑accelerated, integrated)
• Traditional Industrial Automation: ✘ (separate tools)
• Pure‑Play Fusion Labs: ✔︎ (often bespoke)

By bundling these capabilities, Fusionality reduces the engineering headcount required to bring a control system from concept to commissioning by an estimated 30‑40 %, according to internal benchmarks shared with investors.

Funding, Team & Milestones

• Pre‑seed round: $3.7 M (CHF 3 M) led by Founderful and Playfair, closed in Q2 2026.
• Team: 7 full‑time engineers and scientists, including two former DeepMind research engineers, an EPFL plasma‑physics postdoc, and senior hardware designers from the aerospace sector.
• Milestones achieved:
  1. Prototype validation on EPFL’s Tokamak à Configuration Variable (TCV) testbed – demonstrated stable plasma control for 0.5 s shots using the Fusionality stack.
  2. Digital twin release (v1.0) – open‑source core library with a permissive MIT license, already forked by three external fusion groups.
  3. First commercial contract signed with Realta Fusion for a pilot deployment on their compact spherical tokamak, slated for Q1 2027.

The founders stress that the company’s growth strategy is deliberately lean: rather than scaling a large manufacturing operation, Fusionality partners with established PCB assemblers and power‑electronics manufacturers, focusing its internal resources on software, system integration and AI model development.

Read the full breakdown originally published at https://ltdeveloperblogs.github.io/posts/google-deepmind-alumni-are-building-tools-to-accelerate-fusion-power-for-the-grid/