Energy and IT Technology in 20 Years:A Prediction Based on Current Research
Progress
Alfred HüblerSanta Fe Institute
andCenter for Complex Systems Research
University of Illinois at Urbana-Champaign
Predicted Technological Breakpoints: -Merger of information and energy devices (Objective of DOE Smart Grid Initiative)
- Innovation driven by ANN which use humans and mixed reality
Current Research Progress: - Digital Batteries (material with highest energy density & power density, inexpensive, nano)
-Digital Wires (robust power distribution & storage, move and process information)- Atomic Neural Nets (nano-scale particle swarms, which self-assemble into fractal patterns,
which detect patterns, make abstractions; which innovate by association; which exceed computational capacity of humans by a factor of 109, and need less power)
Digital BatteriesAlfred W. Hubler and Onyeama OsuagwuCenter for Complex Systems Research, UIUC
Digital batteries are arrays of nano junctions: - where charge recombination is quantum- mechanically forbidden- where each capacitor can be individually charged/discharged, as in a flash drive- where design prevents tunneling, even if the energy density is very high- which can be integrated on the wafer with sensors, CPUs- which have an energy density > 1 GJ/m3 (200 kJ/kg), charging-discharging rates in the THz range, and exceed number of charging cycles of chemical batteries and conventional capacitors by orders of magnitude.- which are fully operational in a large temperature range (from -273oC to 500oC) and have no thermal run-away
We find:- main problem: SiO2 compressive strength of 1 GPa limits energy density to 200 kJ/kghttp://www.physics.uiuc.edu/people/Hubler/ http://server10.how-why.com/blog/
Energy storage in conventional capacitorsCapacitors are environmentally friendly, work in a large temperature range (0K-melting temperature of metal ), and have a virtually unlimited number of charging cycles.
The energy stored in a capacitor is: W = ½ C V2 , (1)where C=ε A / d is the capacity,
V = applied voltage, ε = electric constant, A = plate area, d = plate distance
The energy density is: w = ½ ε E2 , (2) Where the electric field, E = V/d
However, if the energy density in conventional capacitors exceeds E=3 x 106V/m in air (6 x
107V/m in Teflon) the capacitor discharges by arcing and the energy is lost. => Theoretical value of maximum energy density is small,
w = 100 KJ/m3 (500J/kg) Conventional capacitors need a long time (t ~ ) to charge/discharge since inductance
L is large.
Energy storage in chemical batteries, hydrogen fuel cells, and gasolineEnergy stored in chemical systems is stored as electrostatic energy, as in capacitors. But, in chemicals such as hydrogen, the limiting electric fields are much higher. Quantization phenomena at the atomic level prevent charge recombination => high energy density. Atomic hydrogen is a good example. Energy could be stored in a hydrogen atom by lifting the electron from the ground state to the highest excited state (ionization). In this case, the ratio between the stored energy and the volume of the atom is
w = 13.6eV / (volume of hydrogen atom) = 3.3 x 1013J / m3 (1.31 x 1012J/kg)
i.e. nine orders of magnitude above the maximum energy density in a conventional capacitor. Since the excited state of hydrogen atoms is short lived, hydrogen atoms cannot be used for long term energy storage. For this reason, hydrogen molecules—and carbohydrates, such as gasoline—are commonly used for energy storage. Unfortunately, molecular hydrogen is difficult to handle and the energy retrieval from hydrogen and carbohydrates in fuel cells is slow and inefficient, works only in a small temperature range, and experimental energy density << limit.Energy storage in faradic systems has low efficiency and is limited by diffusion, reaction rates, fractal growth & irreversible chemical reactions.
Digital batteriesDigital batteries are arrays of nano vacuum tubes = arrays of nano vacuum capacitors.
Digital batteries are arrays of nano-scale junctions , where
field emission, avalanche breakdown and Zener breakdown are prevented by quantization phenomena, and which are similar to:-LEDs and laser diodes, but without charge recombination or tunneling,
-Magnetic tunneling junctions, but much simpler in design and cheaper to build
This work builds on our Correlation Tunnel Device patent [H. Higuraskh, Toriumi, F. Yamaguchi, K. Kawamura, A. Hübler, Correlation Tunnel Device, A.U. S. Patent # 5,679,961 (1997)]
Break down probability versus junction size (Alpert et al, Boyle t al., Hubler et al.)
Digital battery
Digital batteriesWe find: Nano vacuum capacitors arrays could sustain energy densities up to 10MJ/kgwithout significant charge recombination,
however the compressive strength of the materials (1GPa for SiO2) limits the energy density to
Emax = compressive-strength / density
= 200 kJ/kg (for SiO2 substrates)
The charge – discharge rate is limited by the induction
f = junction-size / speed-of-light
which is in the THz range.The energy density of chemical batteries is less than 1 kJ/kg.The charge – discharge rate of batteries is limited by diffusion and reaction rates.
Digital batteries are similar to nano plasma tubes, except that
they store energy instead of converting it to light
Digital batteries: Power Density and Energy Density
Fast and light Small and light
Digital batteries are arrays of nano vacuum tubes = arrays of nano vacuum capacitors.
Christopher L. Magee, Massachusetts Institute of Technology,”Towards quantification of the Role of Materials Innovation in overall Technological Development”, http://cmagee.mit.edu/images/docs/chfquantificationofmaterialsrolea.pdf
Nano-junction arrays as Digital Batteries
One could design large arrays of individually connected nano-junction, which could be charged and discharged one-by-one,
similar to flash drive technology.
In contrast to conventional batteries, the output voltage would remain constant until the last nano-capacitor is discharged and
charging/discharging digital batteries would be orders of magnitude faster. Such arrays of nano-capacitors could serve
as digital batteries.
Digital batteries would produce a stable output voltage, making them ideal for sensors and other sensitive devices.
Digital batteries could be recharged probably millions of times, whereas chemical batteries can be recharged only a few
thousand times.
Digital batteries are similar to flash
drives: flash drives store charge, while
digital batteries store energy
ConclusionDigital batteries are potentially an inexpensive and environmentally-friendly alternative to both chemical Batteries.Digital batteries are arrays of nano junctions: - where charge recombination is quantum-mechanically forbidden- where each capacitor can be individually charged- discharged, as in a flash drive- where design prevents tunneling, even if the energy density is very high- which can be integrated on the wafer with sensors, CPUs- which have high energy density, up to 1 GJ/m3 (200 kJ/kg), charging-discharging rates in the THz range, and exceed number of charging cycles of chemical batteries and conventional capacitors by orders of magnitude.- which are fully operational in a large temperature range (from -273oC to 500oC) and have no thermal run-away- main problem: SiO2 compressive strength = 1 GPa (200 kJ/kg) http://www.physics.uiuc.edu/people/Hubler/ http://server10.how-why.com/blog/
Digital batteries are similar to flash drives:
flash drives store charge, while digital batteries store energy
Digital WiresAlfred Hubler, [email protected], Physics, UIUC http://server10.how-why.com/blog
Analog wires are used to move energy (power lines, power grid) and information (data transmission lines, Internet) in electrical networks. However, most dynamical systems with more than 7 degrees of freedom are chaotic => the dynamics of large networks of analog wires is unstable => congestions & cascading failures
Digital wires: Wires that propagate only patterns of rectangular pulses
Specific advantages of digital wires:- Fixed pulse shape (increased reliability & speed);- Robust against electric smog (increased reliability & speed);- No cross talk (increased reliability & speed);- No echoes (increased reliability & speed);- Adjustable pulse speed (increased adjustability);- Encryption (increased security);- Digital wire can be general purpose computers (increased adjustability).Neurons are digital wires. Digital wires move information in parallel.
Digital Wire
Graphical depiction of a “Digital Wire” formed by imposing circular boundary conditions on a CA. A digital wire implemented by a CA has
notable advantages over copper wires, such as isolating defects. By choosing the appropriate rule, the effects of a defect such as a short
circuit can be isolated to a single cell site, rather than propagating the defect along the entire wire, as copper wires do.
Digital wires
Digital Wires: hardware implementation as a transistor network
Vin Vout
Cell 1 Cell 2 Cell n
(a) A digital wire constructed of resistors and pnp transistors. (b) Experimental measurement of the input-output response of the pnp-
transistor digital wire. The sharp transition at 77% of the total supply voltage results in a noise immunity threshold. In order for noise to affect
the outcome of a signal, it must exceed this threshold.
(a) (b)
Digital Wires:
Hardware implementation as a Boolean network
Digital wires: a simple model
Definition: A digitial wire is a long network of cells. Digital pulses travel along the digital wire,
according to the following rule:
Ax+1,y = f (Ax,y-1, Ax,y, Ax,y-1)
-i.e. the state of the cell Ax+1,y =0,1, depends only on the “upstream” neighbors.
Discussion: Digital wires can be viewed as hardware implementations of elementary
cellular automata (S. Wolfram). Therefore a digital wire can be a general purpose computer.
Digital wire (Boolean network, xor rule)
Digital Wire
Direction ofpulse
propagation
Data Program
Digital Wire
Digital Wire
Digital Wire
Digital Wire
Digital Wire
Discussion, continued …Digital wires on various scales:-nano- level: thin film transistor networks (parallel , reliable input for CPUs, may replace CPU), quantum dot networks, neurons (brain)-Atomic level: electron hopping from atom to atom along a path on a macro molecule (hard ware implementations of neural nets)-Microscopic level: transistor networks-Mesoscopic level: Boolean networks, Field programmable gate arrays (image processing)- Macroscopic level: power lines with phase sensitive switches every 10 miles (no cascading power failures), city trafficData transmission lines versus power lines:There is energy traveling with every pulse. Computation does not necessarily consume much power (conservative computation). -Periodic pulses can produce a lot of power. -Pulses that carry information look random.H. Higuraskh, A. Toriumi, F. Yamaguchi, K. Kawamura, A. Hübler, Correlation Tunnel Device, U. S. Patent # 5,679,961 (1997)
Digital Wire
Discussion, continued …Different cellular automata rules:-Rule 110: general purpose computer-Rule 204: identity rule-Rule 30: random number generator-Rule 254: self-repairing pulses-Rule 0: trivial
Merging data from different digital wires:
Digital Wire
Wire 1 Wire 2
Given is the state 000111010.
What is the pulse one time step later for rule 0?
000000000
Summary: Digital WiresAnalog wires are used to move energy (power lines, power grid) and information (data transmission lines, Internet) in electrical networks. -Dynamical systems with more than 7 degrees of freedom are chaotic (Lee Rubel )=> the dynamics of large networks of analog wires are unstable.
Digital wires: wires that propagate only patterns of rectangular pulses (thresholds)
Specific advantages of digital wires:- Fixed pulse shape (increased reliability)- Robust against electric smog (increased reliability)- No cross talk (increased reliability)- No echoes (increased reliability)- Adjustable pulse speed (increased adjustability)- Encryption (increased security)- Digital wire can be general purpose computer s(increased adjustability)
Human Neurons are digital wires.Alfred Hubler, [email protected], Physics, UIUC http://server10.how-why.com/blog
Digital Wire
Atomic Neural Nets: Self-assembly of a particle swarms into wire networks with thresholds.
Experiment: Agglomeration of conducting particles in an electric field1) We focus on the dynamics of the system2) We explore the topology of the networks using graph theory.3) We explore a variety of initial conditions.
random initial distribution compact initial distribution
Atomic Neural Nets: Description of experimental setup
Basic experiment consists of two electrodes, a source electrode and a boundary electrode connected to opposite terminals of a power
supply.
source electrod
e
boundary electrode
battery
Atomic Neural Nets: Description of experimental setup
Basic experiment consists of two electrodes, a source electrode and a boundary electrode connected to opposite terminals of a power
supply.
The boundary electrode lines a dish made of a dielectric material
such as glass or acrylic.
The dish contains particles and a dielectric medium (oil)
source electrod
e
boundary electrode
oil
battery
particle
Atomic Neural Nets: Description of experimental setup
20 kV
battery maintains a voltage difference of 20 kV between boundary and source
electrodes
Atomic Neural Nets: Description of experimental setup
20 kV
source electrode sprays charge over oil surface
Description of experimental setup
20 kV
source electrode sprays charge over oil surface
air gap between source electrode and oil surface approx. 5 cm
Atomic Neural Nets: Description of experimental setup
20 kV
source electrode sprays charge over oil surface
air gap between source electrode and oil surface approx. 5 cm
boundary electrode has a diameter of 12 cm
Atomic Neural Nets: Description of experimental setup
20 kV
needle electrode sprays charge over oil surface
air gap between needle electrode and oil surface approx. 5 cm
boundary electrode has a diameter of 12 cm
oil height is approximately 3 mm, enough to cover the particles
castor oil is used: high viscosity, low ohmic heating, biodegradable
Atomic Neural Nets: Description of experimental setup
20 kV
needle electrode sprays charge over oil surface
air gap between needle electrode and oil surface approx. 5 cm
ring electrode forms boundary of dish
has a radius of 12 cm
oil height is approximately 3 mm, enough to cover the particles
castor oil is used: high viscosity, low ohmic heating, biodegradable
particles are non-magnetic stainless steel, diameter D=1.6 mm
particles sit on the bottom of the dish
Phenomenology Overview
12 cm
t=0s 10s 5m 13s 14m 7s
stage I:strand
formation
Phenomenology Overview
12 cm
t=0s 10s 5m 13s 14m 7s
14m 14s
stage I:strand
formation
stage II:boundary
connection
Phenomenology Overview
12 cm
t=0s 10s 5m 13s 14m 7s
14m 14s 14m 41s 15m 28s
stage I:strand
formation
stage II:boundary
connection stage III: geometric expansion
Phenomenology Overview
12 cm
t=0s 10s 5m 13s 14m 7s
14m 14s 14m 41s 15m 28s 77m 27s
stage I:strand
formation
stage II:boundary
connection stage III: geometric expansion
stationary state
Motion of the strands: pointed equilibrium
The motion of the lead particles of the six
largest strands from a single experiment.
Adjacency defines topological species of each particle
Termini = particles touching only one other particle
Branching points = particles touching three or more other particles
Trunks = particles touching only two other particles
Particles become one of the above three types in stage II and III. This occurs over a relatively short period of time.
Relative number of each species is robust
Graphs show how the number of termini, T, and branching points, B, scale with the total number of particles in the tree.
Most networks are trees.Only a few rare cases contain loops
(cycles).
Loops (cycles) are unstable
Insets on the left show two particles artificially placed into a loop separate from one another.
The graph on the right shows the separation between the two particles as a function of time.
Fractal Dimension
Particles arrange themselves similarly in different experiments.
Overall electrical resistance of system
The resistance decreases as a function of time. The limiting value is reproducible. If the current is fixed, the system minimizes energy consumption.
Predicting Network Growth: Qualitative effects of initial distribution
Qualitative Predicting Network Growth: Qualitative effects of initial distributions of initial distribution
N = 752T = 131B = 85
N = 720T = 122B = 106
N = 785T = 200B = 187
N = 752T = 149B = 146
Initial conditions have a strong influence on the number of trees and are a strong constraint on the final form of tree(s).
Qualitative Predicting Network Growth: Qualitative effects of initial distribution
Will this initial configuration produce a spiral?
?
Qualitative Predicting Network Growth: Qualitative effects of initial distribution
No, system is unstable to ramified structures.
Qualitative Predicting Network Growth
Since topology of the networks is established relatively quickly, particles connect to one another before they have moved far.
Thus, we attempt to model the connections formed by the system using only the local information for each particle—it’s neighborhood.We use data from the experiments: a snapshot of the particles directly preceding stage II.
Qualitative Predicting Network Growth
Since topology of the networks is established relatively quickly, particles connect to one another before they have moved far.
Thus, we attempt to model the connections formed by the system using only the local information for each particle—it’s neighborhood.We take data from the experiments: a snapshot of the particles directly preceding stage II.
Digitize the positions.
Run the adjacency algorithm to obtain a base neighborhood.
cutoff length = 3 particle diameter
Predicting Network Growth: Sequences of disruptions with different likelihood
Growth models:
Particles articles can only connect to particles that neighbor it.
Algorithms run until all available particles connect into a tree.
Some particles will not connect to any others (loners). They commonly appear in experiments.
We chose three growth models:
1) random growth model: all neighbors equally likely to connect, but no loops
2) minimum spanning tree model: closer neighbors a more likely, no loops
3) propagating front model: one neighbor has to be connected, no loops
loner
loner
Predicting Network Growth: Random Growth Model
Typical connection structure from RAN algorithm.
Distribution of termini produced from 105 permutations run on a single
experiment.
Number of termini produced for all experiments, plotted as a function of
N.
Predicting Network Growth: Minimum Spanning Tree Growth
Typical connection structure from MST algorithm.
Distribution of termini produced from 105 permutations run on a single
experiment.
Number of termini produced for all experiments, plotted as a function of
N.
Predicting Network Growth: Propagation Front Model
Typical connection structure from PFM algorithm.
Distribution of termini produced from 105 permutations run on a single
experiment.
Number of termini produced for all experiments, plotted as a function of
N.
Comparison of all models to experiments
The number of termini and branching points for all three models and the natural experiments.
The minimum spanning tree model produces the most accurate prediction of the experimental data.
Predicting the growth of a fractal particle network.
Experiment: J. Jun, A. Hubler, PNAS 102, 536 (2005)1) Statistically robust features: number of termini, number of branch points,
resistance, open loop, Three growth stages: strand formation, boundary connection, and geometric expansion;
2) Features that depend sensitive on noise, initial conditions and other external influences: number of trees, ….
3) Minimum spanning tree ensemble predictor predicts emerging pattern best therefore these self-assembling, self-repairing networks could be used as ensemble predictors.
Applications: Hardware implementation of neural nets, nano neural nets with SC particles - M. Sperl, A Chang, N. Weber, A. Hubler, Hebbian Learning in the Agglomeration of Conducting Particles, Phys.Rev.E. 59, 3165 (1999)
random initial distribution compact initial distribution
Hebbian Learning in a three-electrode system:Pattern recognition, abstraction, innovation by
association …
M. Sperl, A Chang, N. Weber, A. Hubler, Hebbian Learning in the Agglomeration of Conducting Particles, Phys.Rev.E. 59, 3165 (1999)
Energy = positive experience
Atomic Neural Nets: Basic Units
Figure: A self-assembling wires unit interacting with a virtual environment.
Energy = positive experience
Atomic Neural Nets: Pre-wired Networks of Basic Units
Figure: A network of basic units with nonlinear input nodes with a threshold. The lines indicate pre-wired connections. Sub-network may emerge, when the self-assembling wires use certain pre-wired connections and ignore others.
Atomic Neural Nets
Experiments by Peter Fleck et al. show that superconducting nano- particles behave similarly.
The wires of such atomic neural nets, have a diameter of roughly 1 nanometer, whereas human neurons have a diameter of roughly 1 micrometer. Therefore:
-1 billion atomic neural net neurons fit have the same volume as one human neuron- the power consumption of these 1 billion atomic neural net neurons is less than that of one human neuron- the behavior of atomic neural net neurons, depends on materials, geometries, …
Conclusion: The number of neurons in Atomic Neural Nets can exceed number of neurons in human brains by a factor of 109 and use less power.
Levels of Understanding of Perceptrons (=Machine with understanding)
Atomic neural nets may reach a level of understanding that is incomprehensible for humans
… and speak and read English (such as the How-Why tutoring system)
Understanding = ability to translate -between observations and a conceptual network (virtual world)-between conceptual networks
Energy and IT Technology in 20 Years:
A Prediction Based on Current Research Progress
Current Research Progress: - Digital Batteries (material with highest energy density & power density, inexpensive, nano-scale)- Digital Wires (robust power distribution & storage, move and process information)- Atomic Neural Nets (nano-scale particle swarms, which self-assemble into fractal patterns, which detect patterns, make abstractions; which innovate by association; which exceed computational capacity of humans by a factor of 109, and need less power) Predicted Technological Breakpoints: - Merger of information and energy devices - Innovation driven by self-assembling ANN which ‘understand’ the world better than humans