Track 03 / Thermodynamic Computing
Thermodynamic Computing
Probabilistic hardware, energy-based models and what a bit costs in joules.
- Duration
- 4 weeks
- Pace
- Self-paced
- Modules
- 8
- Live sessions
- 8, optional
- Price
- $24.99 founding, $49.99 later
- Seats
- Limited to 25
Starts in December and runs into the first week of January, with a break over the holidays.
Free to join. Nothing to pay now.
01 / The climb
More computation per joule.
K = 0.73
02 / Who it's for
Who it's for
- 01Hardware, physics and systems engineers who want to see computing without deterministic logic.
- 02Machine learning people who want to understand sampling and energy-based models properly.
- 03Researchers asking how far AI's energy costs could fall.
03 / Prerequisites & time
Prerequisites and time
- 01Python and some numerical computing.
- 02Probability and linear algebra at first-year university level.
- 03Introductory physics helps. Statistical mechanics is taught from the beginning.
- Duration
- 4 weeks
- Pace
- Self-paced
- Note
- Starts in December and runs into the first week of January, with a break over the holidays.
04 / Syllabus
Four weekly blocks
Week 1
M01
Entropy, information and heat
Statistical mechanics essentials: microstates, entropy, free energy and the Boltzmann distribution.
Proof of work
A simulation of a small system that recovers its Boltzmann distribution and entropy numerically.
M02
Landauer's principle and the cost of computation
The minimum energy to erase a bit, reversible computing, and how far real hardware is from the bound.
Proof of work
A worked estimate of energy per operation for a chosen device, compared with the Landauer limit.
Week 2
M03
Stochastic computing and sampling
Random bits as a primitive, p-bits, stochastic circuits, Metropolis–Hastings, Gibbs sampling, mixing times and annealing.
Proof of work
A stochastic circuit simulation and samplers for a target distribution, with convergence diagnostics and error versus sample count.
M04
Energy-based models
Energy functions, partition functions, contrastive learning and score-based views.
Proof of work
An energy-based model trained on a small dataset, with generated samples and a training report.
Week 3
M05
Ising models and Boltzmann machines
Spin systems, restricted Boltzmann machines and learning with sampling.
Proof of work
A Boltzmann machine trained from scratch, with weights checked against a known target distribution.
M06
Sampling hardware
Physical samplers, noise sources, analogue and probabilistic devices, and their limits.
Proof of work
A hardware-aware sampler model with noise and precision constraints and a sensitivity analysis.
Week 4
M07
Connections to AI efficiency
Energy per inference, where sampling dominates cost, and what hardware could change.
Proof of work
A comparative model of energy cost for a sampling-heavy workload on conventional and probabilistic hardware.
M08
Capstone: a thermodynamic workload
Select a sampling or optimisation problem and study how it would run on probabilistic hardware.
Proof of work
A reproducible study with an energy analysis, reviewed with the platform and, if you want, in a live session.
05 / Format
Format
01
Self-paced
Four weeks. Work whenever suits you.
02
Eight optional live sessions
A kickoff session and seven more with a human expert in the field. You can come to all of them, some, or none.
03
Small cohort
Limited to 25 people per course.
06 / Outcomes
Outcomes
- 01A sense of the physical limits on computation, and how far today's hardware is from them.
- 02Working samplers, and energy-based models and Boltzmann machines you trained yourself.
- 03A model of a probabilistic hardware design with realistic noise and precision limits.
- 04A way to put numbers on claims about energy-efficient AI.
07 / FAQ
FAQ
Is thermodynamic computing a product?
Here it is a field of study: the physics of computation, and hardware that uses randomness directly. The track is not tied to any vendor.
Do I need a physics degree?
No. The first module teaches the statistical mechanics you need.
How does it relate to the AI / SI track?
They share ideas about sampling and energy-based models. You can take either one alone.
When does it start?
The first cohort starts in December and runs into the first week of January, with a break over the holidays. It is limited to 25 people.
03 / Thermodynamic
Reserve your place
Free to join. Nothing to pay now. Founding price $24.99 per course. Later cohorts are $49.99.