Compute that closes
its own thermodynamic budget.

In vacuum, heat is only what you can radiate. ThermoEdge designs the compute to the limit that forces the question. The same core deploys in every easier envelope.

One number prices every decision.

min   J/opcompute  ·  J/bitcomms
joules per useful operation · joules per delivered bit

Every architectural choice is priced by this product. Lower it — the choice is in. If a choice does not lower it, it does not ship.

In vacuum the metric sharpens, because heat is not a standalone cost. Heat is the head of a mass cascade: every watt drags a radiator, an array, a coolant loop, a propellant tank, and the launch mass to lift all of it. Cutting joules-per-op multiplies down the entire spacecraft budget, not just the thermal subsystem.

Attack the joule-sinks largest first.

Thermodynamic optimization is not chasing the Landauer floor. It is removing absolute dissipation, in descending order of magnitude, with the floor as the asymptote the cost surface measures against.

01
largest sink

System & architecture overhead

Data movement between memory and compute. Idle and standby. The emulation tax of running math on a general-purpose stack. And, on any link, the cost of transmitting a bit.

Architectural answer

In-memory compute (no movement). Event-driven operation (no idle). A microkernel with math as hardware primitives (no emulation). Decisions-not-data downlink (no needless transmission).

02
the device-level sink

Dynamic switching — CV²

Every conventional gate charges and discharges a capacitor each cycle. The charge becomes heat. This is the largest order-of-magnitude lever available before approaching the floor.

Architectural answer

Adiabatic / reversible logic recovers a fraction of the switching energy through resonant clocking. Recovered energy is energy never radiated.

03
the floor under everything

Static / standby leakage

Conventional memory leaks power to retain state. DRAM refreshes. SRAM holds. A node that sleeps still draws.

Architectural answer

Unified persistent memory: non-volatile under the address space. A power cycle preserves every byte. Sub-mW idle. No refresh, no retention tax.

04
the asymptote

The Landauer floor — kT·ln2

The thermodynamic lower bound for erasing one bit. Real silicon today runs roughly 10⁹–10¹¹× above it. The floor is a far asymptote, not a near target.

Architectural answer

Logical reversibility, in principle. Approached only after the first three rungs are largely solved. The cost surface measures distance to this floor and minimizes absolute dissipation against it.

In vacuum, every watt drags everything.

The only way to shed heat from a spacecraft is to radiate it. The Stefan–Boltzmann law sets the rate, and every term in the spacecraft budget scales with it.

Radiative cooling — the only mechanism

Prad  =  ε·σ·A  ( T4  −  Tspace4 )
ε emissivity · σ Stefan–Boltzmann · A radiator area

The rate goes as the fourth power of temperature. A part rated for 100 °C junction sheds heat at well over 2.5× the rate of one held at room temperature, on the same area. Running the silicon hot makes the radiator small. The single largest lever on radiator area — and one ThermoEdge owns, because the silicon is ours.

Design choice: high-junction-temperature silicon (≥90–100 °C)

A watt of compute drags the whole spacecraft

+1 W compute becomes waste heat → radiator area scales by 1 / T⁴ headroom → solar array to generate that watt → coolant + pumps to move the heat → pump power to drive the coolant → launch mass to lift all of the above CUT HERE

Cut the compute joule and the entire cascade shrinks. The demand-side lever is worth more in vacuum than anywhere else — precisely because the supply-side tail is longest there.

Transmitting a bit costs more than computing one.

J/bitlink  ≫  J/opcompute
across most link budgets, by orders of magnitude

The link budget is a physical law before it is an engineering choice. Free-space loss, channel capacity, and antenna gain put a hard joule price on every delivered bit. On a small spacecraft, the comms term is frequently the single largest dissipation source.

The architectural response: never transmit what you can avoid. Compute the inference at the data source. Emit the decision, not the raw stream. Wake on event, sleep otherwise. Compute-at-edge is a comms-energy lever first, a latency lever second.