The full case for a universal substrate.

For more than seven decades, every major advance in computing has been forced to speak an old language.

We built extraordinary new machines — photonic processors that compute with light, neuromorphic chips that fire like neurons, quantum systems that exist in superposition — and then demanded they conform to the architecture and software models of classical digital computers.

The result is fragmentation.

Powerful specialized hardware remains isolated. Software becomes a translation layer instead of a native medium. Progress is slowed by the friction of incompatibility.

The deepest bottleneck in computing is no longer only transistor density or energy efficiency. It is the absence of a shared substrate — a common representational layer that can treat every form of computation as a first-class citizen.

HeteroIR exists to end that absence.

HeteroIR is a living, graph-based intermediate representation designed from the ground up for heterogeneous computation.

It is not another compiler IR that begins with classical assumptions and then tries to accommodate everything else. It is a neutral, expressive substrate in which classical digital operations, photonic transformations, neuromorphic event dynamics, and quantum processes can coexist as equal citizens of the same graph.

• Computation is expressed as dynamic graphs of operations and relationships.
• Different physical substrates are mapped onto the graph rather than the reverse.
• Data movement, timing, probability, continuity, and discrete events are all native concepts.
• New forms of computation can be added without breaking the foundation.

HeteroIR does not merely connect different kinds of hardware. It gives them a shared world in which to operate.

When every new computing paradigm requires its own software stack, its own tools, and its own experts, progress remains siloed. A universal substrate changes the economics and the speed of innovation:

• Hardware designers can focus on physical excellence without simultaneously inventing a complete software ecosystem.
• Software and algorithm designers can target capabilities rather than specific machines.
• Systems can fluidly combine the strengths of multiple paradigms in a single workload.
• The arrival of future computational substrates becomes an addition to the graph rather than a rupture.

This is how computation evolves from a collection of isolated tools into a coherent medium.

The true significance of HeteroIR is not technical elegance. It is what becomes possible when the friction between different forms of intelligence is removed.

Planetary systems can be modeled with continuous fidelity. Biological processes can be simulated at meaningful scales. Materials, energy systems, and complex optimization problems can be explored as ongoing processes rather than rare, heroic efforts.

When the architecture of any single machine is no longer the limiting factor, the limiting factor becomes the quality of the questions we ask.

HeteroIR is infrastructure for that shift.

HeteroIR is being developed as open foundational infrastructure. The work includes:

• A formal specification of the intermediate representation
• Reference compilers and orchestration systems
• Simulation environments that allow development before physical hardware exists
• Runtime systems capable of managing truly heterogeneous execution
• Collaboration with hardware, software, and research communities

This is not a product in the conventional sense. It is an attempt to supply the missing layer that heterogeneous computing requires.