← Back to PMBT main site
Continuous Bio-Computation

Analog Bio-Computers
Composed of Trinary Processors

Beyond the discrete −1 / 0 / +1 trits of a single PMBT chip lies a second, richer regime: when many trinary biochemical processors are wired together by microfluidic channels, the metal-ion concentrations flowing between them become continuous real-valued signals. The whole array stops behaving like a digital circuit and starts behaving like an analog computer — one that solves equations directly in the chemistry of its enzymes.

Each processor becomes an analog processing element whose transfer function is its enzyme kinetics. Summation happens by mixing, multiplication by coupled cascades, integration by accumulation, and differentiation by feed-forward sensing. Thousands of these elements form a reaction–diffusion fabric that natively integrates differential equations in physical time.

View the Digital Processor
N × N
Processor mesh
Continuous domain
6
Analog operators
ODE / PDE
Solved in chemistry
01 / Paradigm

From discrete trits to continuous signals

A single PMBT processor quantises each metal channel into three logic states. But the underlying chemistry never truly leaves the continuum — concentrations, reaction rates and reporter intensities are smooth real numbers. An analog computer simply refuses to quantise: it keeps the full real-valued signal and computes with it directly.

PropertyDigital PMBT (trinary)Analog array (continuous)
State spaceDiscrete { −1, 0, +1 }Continuous concentration [M] ∈ ℝ⁺
Information carrierOperon switched on / partial / offReal-valued metal-ion flux & reporter intensity
Elementary unitTrit (1.58 bits)Analog level (precision-limited by noise)
Composition lawBoolean / trinary logic gatesSuperposition, Kirchhoff-like mass balance
Time modelClocked transitionsContinuous real-time dynamics
Natural problemDecision, classification, logicDifferential equations, optimisation, filtering
Why analog, why now

Detoxification kinetics are inherently graded — the rate of conversion scales smoothly with metal concentration. Quantising throws this richness away. Analog mode exploits it.

The trade

Analog gains massive parallelism and native calculus, but pays in precision: each value is limited by molecular noise and cross-talk to roughly 4–6 effective bits per element.

Hybrid sweet spot

PMBT arrays can run digital trinary logic for control and decisions, while dedicating sub-arrays to analog computation — a mixed-signal bio-computer.

02 / Building Block

The Analog Processing Element

Strip a PMBT processor down to a single metal channel and you have an Analog Processing Element (APE): a biochemical transfer function that maps an input concentration to an output signal through enzyme kinetics. Every analog computer on this page is built by wiring APEs together.

Transfer function
// Michaelis–Menten APE
v([S]) = Vmax · [S] / (Km + [S])
// Cooperative / Hill APE (gain g)
v([S]) = Vmax · [S]n / (Kn + [S]n)

Input = metal-ion concentration [S]. Output = product / reporter flux v. The saturating curve gives a built-in soft limiter; the Hill exponent n sets analog gain and turns an APE into a tunable amplifier or comparator.

InputsMetal-ion concentrations delivered by microfluidic channels
WeightsEnzyme expression level, Km, and channel cross-section
NonlinearitySaturation (Vmax) + cooperativity (Hill n)
OutputProduct concentration, fluorescence, or downstream flux
MemoryProduct accumulation gives a natural integrator state
03 / Math in Chemistry

Six analog operations, realised biochemically

A classical analog computer is built from a small set of operators — adders, multipliers, integrators, differentiators. Metabolic chemistry offers a direct biochemical analogue for each, so a network of APEs can assemble these primitives into arbitrary computations.

Σy = Σ aᵢ · xᵢ
Summation
Parallel reactions in a shared compartment

Several APEs releasing the same product into one micro-chamber add their fluxes by simple mass balance — the analog sum of their inputs.

×y ∝ x₁ · x₂
Multiplication
Coupled / cascaded enzymatic stages

When one channel’s product is the substrate of another, output scales with the product of both inputs — an analog multiplier built from sequential cascades.

y(t) = ∫ x(τ) dτ
Integration
Product accumulation over time

A sealed chamber that accumulates reaction product holds the time-integral of its input flux — the core element for solving differential equations.

d/dty ∝ dx/dt
Differentiation
Incoherent feed-forward loop

An activator racing a delayed repressor produces a transient proportional to the rate of change of the input — a biochemical differentiator.

logy ∝ log(x), eˣ
Log / Exp
Cooperative (Hill) binding

Steep cooperative response curves approximate logarithmic and exponential maps, enabling analog companding and wide dynamic range.

y = 1 if x ≥ θ
Threshold / Compare
Ultrasensitive bistable switch

A zero-order ultrasensitive motif acts as an analog comparator, snapping output high once an input crosses a tunable set-point.

04 / Active Element

The biochemical operational amplifier

The op-amp is the workhorse of electronic analog computers. Its biochemical twin is an enzymatic cascade with high cooperative gain wrapped in product-inhibition feedback. Closed around different feedback chemistries, this single motif becomes an adder, an integrator, a comparator or a buffer — exactly as in silicon.

[S]+refcascadegain Aoutproduct inhibition
Open-loop gainHigh, from cooperative enzymatic cascades (Hill n > 1)
Negative feedbackProduct inhibition closes the loop and linearises response
Virtual set-pointAllosteric reference holds an effective “virtual ground”
Saturation railsVmax defines the upper / lower output rails

Configurable by feedback: capacitive feedback (an accumulation chamber) → integrator; resistive feedback (a leak channel) → weighted adder; no feedback → comparator.

05 / Architecture

The reaction–diffusion processor fabric

Tile APEs onto a 2D (or stacked 3D) grid and connect neighbours with microfluidic channels. Substrate diffusing between cells couples them spatially; the array as a whole becomes an analog computer whose state evolves according to a reaction–diffusion equation — computing a field, not a number.

8 × 8 tile of trinary processors — each cell an analog element, edges = diffusive coupling

// governing dynamics of the fabric
∂[M]/∂t = D∇²[M] + R([M])
D = diffusion (channel geometry)   R = local APE reaction term
Massive parallelism

Every cell computes simultaneously — the array updates as one continuous physical system, no clock required.

Spatial coupling

Channel widths set diffusion weights, turning the mesh into a programmable analog interconnect.

Emergent computation

Pattern formation, wave propagation and gradient following emerge directly from the chemistry.

Scalable

Tiles compose into wafers; multiple metals give independent, superimposed computing layers.

06 / What it computes

Problems the analog array solves natively

Because the hardware obeys the same equations as the problems, the analog bio-computer does not simulate these systems — it physically embodies them and relaxes to the answer.

ODE
Ordinary differential equations

Chain integrators and adders in feedback to model exponential decay, oscillators and chemical-kinetic systems in real, physical time.

PDE
Reaction–diffusion PDEs

The fabric natively integrates ∂u/∂t = D∇²u + f(u) — the array IS the discretised PDE, solving it as it physically evolves.

OPT
Optimisation & gradient descent

Chemotactic gradient following lets the array roll downhill on a concentration landscape, settling into minima — analog optimisation.

NN
Analog neural inference

Weighted summation + saturating nonlinearity is exactly a neuron. Meshes of APEs perform matrix–vector products and activations as a physical neural network.

DSP
Filtering & signal processing

Integrator–feedback loops form low/high-pass filters that smooth or sharpen incoming chemical signals continuously.

LIN
Linear algebra

Crossbar arrays of weighted APEs evaluate matrix–vector multiplication in a single diffusion step — the analog accelerator pattern.

07 / Reference machines

Analog computers built from PMBT processors

Wiring many analog processing elements together yields whole machines, each a biochemical re-imagining of a classic analog computer. These are reference architectures — blueprints for composing tiles into purpose-built continuous solvers.

BDA-3
Biochemical Differential Analyzer

A direct descendant of Bush’s mechanical differential analyzer, rebuilt in wetware. Cascaded bio-integrators wired in feedback solve systems of ODEs in real, physical time.

Building blocks
6–8 integrator tilesCrossbar feedback patch-bayOptical read-out per node
Solves

Linear & non-linear ODE systems, oscillators, chemical kinetics

RDFC
Reaction–Diffusion Field Computer

A 2-D sheet of coupled APE tiles that physically integrates ∂u/∂t = D∇²u + f(u). The substrate IS the discretised PDE — the answer emerges as the field relaxes.

Building blocks
64×64 diffusion latticeTunable coupling channelsTime-lapse fluorescence imaging
Solves

Parabolic PDEs, pattern formation, shortest-path & maze solving

XMM
Crossbar Analog Matrix Multiplier

A grid of weighted APEs where each junction stores a conductance. One diffusion step performs a full matrix–vector product — the in-memory analog accelerator pattern.

Building blocks
128×128 weight crossbarColumn current summersADC sampling fringe
Solves

Matrix–vector products, linear systems, transform kernels

MNN
Metabolic Neural Network

Layers of summing op-amp neurons with saturating biochemical activation. Weighted metabolite fluxes feed sigmoidal gates — a physical multilayer perceptron that infers in one pass.

Building blocks
3-layer perceptron meshSaturating bio-op-amp gatesPlastic (trainable) weights
Solves

Classification, analog inference, associative recall

GDO
Gradient-Descent Optimizer Cell

A chemotactic tile that physically rolls downhill on a concentration landscape, settling into the nearest minimum — analog optimisation with zero iterative software loop.

Building blocks
Gradient-sensing membraneMomentum feedback loopMinimum-latch reporter
Solves

Continuous optimisation, energy minimisation, constraint relaxation

PID
Continuous PID Bio-Controller

Proportional, integral and derivative bio-blocks combined into a closed-loop regulator that holds a chemical set-point against disturbances — control theory rendered in metabolism.

Building blocks
P / I / D op-amp trioSet-point reference cellActuator efflux pump
Solves

Homeostatic regulation, set-point tracking, disturbance rejection

08 / Envelope

Indicative performance envelope

Order-of-magnitude figures for a single-metal analog tile. They trade digital precision for extreme parallelism and near-thermodynamic-limit energy efficiency — the characteristic analog bargain.

Dynamic range
~10³–10⁴

concentration span per channel (nM – mM)

Effective precision
4–6 bits

per analog element, noise-limited

Bandwidth
10⁻³–10⁰ Hz

set by enzyme turnover & diffusion

Element density
10⁴–10⁶ / cm²

micro-chambers per chip

Energy / op
ATP-scale

~10⁻¹⁹ J, near thermodynamic limit

Parallelism
Full-array

all elements evolve concurrently

* Theoretical, exploratory estimates — not measured device specifications.

09 / Applications

Where an analog bio-computer wins

Analog wins wherever the problem is itself continuous, parallel and physical — fields, flows, gradients and learning — and where energy budget matters more than bit-exact answers.

Environmental field computing

A sensor sheet that solves the pollutant diffusion equation in situ, mapping contamination spread in real time.

Adaptive bioremediation control

Closed-loop analog controllers modulate detox enzyme expression proportionally to live metal levels.

Metabolic neural networks

Trainable APE meshes perform low-power analog inference on chemical inputs — a wet neural processor.

Reaction-diffusion pattern design

Programmable Turing-pattern fabrics for biomaterials, tissue scaffolding and self-organising structures.

Continuous optimisation

Physical gradient descent solves routing, allocation and parameter-fitting problems in chemistry.

Hybrid mixed-signal cores

Digital trinary logic supervises analog sub-arrays — decisions in trits, math in concentrations.