# ThermoCog

> ThermoCog is design-stage research: the current artifacts are a whitepaper and architectural design. Software simulation and hardware implementation are planned; no experimental results are claimed.

> Thermodynamic Cognition: a design-stage blueprint for 20-Watt biomorphic AGI. The design investigates physical relaxation and alternatives to backpropagation; these ideas are hypotheses, not demonstrated AGI.

ThermoCog investigates energy and data-movement costs in machine intelligence. Large GPU deployments can consume substantial power; this design supplies no measured universal training load, scaling requirement or data-movement energy fraction. ThermoCog grounds artificial intelligence in non-equilibrium thermodynamics, analog memristive physics, and multi-timescale biological principles.

## Core architecture

- [ThermoCog Architecture](https://thermocog.com/#architecture): Proposed biomorphic architecture; no approach to Landauer bounds is measured
- [Thermodynamic Floor](https://thermocog.com/#thermodynamics): Fundamental physics of Landauer dissipation and the Von Neumann memory wall
- [Three Biological Timescales](https://thermocog.com/#timescales): Tri-level hierarchy spanning phylogeny, ontogeny, and cognition
- [Physical Energy Relaxation](https://thermocog.com/#relaxation): Equilibrium propagation computing loss gradients without backpropagation
- [Neural Cellular Automata](https://thermocog.com/#morphogenesis): Research into local topology development and possible defect tolerance
- [Homeostatic Sleep](https://thermocog.com/#sleep): Proposed offline consolidation; effects on saturation and forgetting remain untested
- [Active Inference Engine](https://thermocog.com/#inference): Minimization of variational free energy across embodied perception-action loops
- [Hardware Realization Matrix](https://thermocog.com/#specs): Component specifications comparing biomorphic silicon against digital accelerators
- [Silicon Roadmap](https://thermocog.com/#roadmap): Progression from deterministic MCU firmware to mixed-signal analog neuromorphic ASICs

## Technical foundations

- [Landauer Limit](https://thermocog.com/#thermodynamics): Physical lower bound of 2.87 zeptojoules per bit erasure at room temperature
- [Von Neumann Memory Wall](https://thermocog.com/#thermodynamics): Research into reducing data movement with in-memory arrays; complete-system costs remain unmeasured
- [Liquid Time-Constant Networks](https://thermocog.com/#timescales): Continuous neural ODEs with adaptive conductance time constants
- [Equilibrium Propagation](https://thermocog.com/#relaxation): Contrastive energy minimization without transposed backward error transport
- [Synaptic Downscaling](https://thermocog.com/#sleep): Global renormalization during sleep restoring dynamic range and signal-to-noise ratio
- [Frequently Asked Questions](https://thermocog.com/#faq): Physical-layer leakage control, sneak-path tiling (1T1R/1S1R), Holomorphic EqProp conductance asymmetry, and commercialization scope
- [Foundational Lexicon & Glossary](https://thermocog.com/#glossary): Definitive glossary of 10 thermodynamic, neuromorphic, and continuous-physics terms

## Primary literature archive

- [Karl Friston (2006)](https://doi.org/10.1016/j.neuron.2006.07.016): A free energy principle for the brain
- [Scellier and Bengio (2017)](https://doi.org/10.3389/fncom.2017.00024): Equilibrium Propagation for training energy-based models
- [Rolf Landauer (1961)](https://doi.org/10.1147/rd.53.0183): Irreversibility and heat generation in the computing process
- [Carver Mead (1990)](https://doi.org/10.1109/5.58356): Neuromorphic electronic systems and analog physical computation
- [Hasler and Marr (2013)](https://doi.org/10.3389/fnins.2013.00118): Finding a roadmap to achieve large neuromorphic hardware systems
- [Tononi and Cirelli (2014)](https://doi.org/10.1093/sleep/37.1.13): Sleep and the price of plasticity through synaptic homeostasis
- [Alexander Mordvintsev et al. (2020)](https://doi.org/10.23915/distill.00023): Growing Neural Cellular Automata for morphogenesis
- [Ramin Hasani et al. (2021)](https://doi.org/10.1609/aaai.v35i9.16937): Liquid Time-Constant networks with input-dependent continuous dynamics

## Author and research context

Cisco Caceres is the author of this design-stage research. [Research portfolio and status](https://ciscocaceres.com/research/#thermocog).

## Canonical links

- [ThermoCog Home](https://thermocog.com/): Official blueprint and interactive specification
- [Full Context for LLMs](https://thermocog.com/llms-full.txt): Comprehensive technical derivations and mathematical models
- [Sitemap](https://thermocog.com/sitemap.xml): Machine-readable index of canonical pages

## Optional


_Generated: 2026-09-19_
