ArticlePDF Available

An Evolved Circuit, Intrinsic in Silicon, Entwined With Physics.

Authors:

Abstract and Figures

. `Intrinsic' Hardware Evolution is the use of artificial evolution --- such as a Genetic Algorithm --- to design an electronic circuit automatically, where each fitness evaluation is the measurement of a circuit 's performance when physically instantiated in a real reconfigurable VLSI chip. This paper makes a detailed case-study of the first such application of evolution directly to the configuration of a Field Programmable Gate Array (FPGA). Evolution is allowed to explore beyond the scope of conventional design methods, resulting in a highly efficient circuit with a richer structure and dynamics and a greater respect for the natural properties of the implementation medium than is usual. The application is a simple, but not toy, problem: a tone-discrimination task. Practical details are considered throughout. 1 Introduction This paper describes a case-study in intrinsic hardware evolution: the use of artificial evolution --- such as a Genetic Algorithm --- to design a circuit automa...
Content may be subject to copyright.
E
F
S
E
S
EN
N
F
S
W
N
F
W
W F
S
EN
W
E
N
S
W
S
N
E
W
F
N
E
S
W
Out
In
configuration
PC
Desktop
XC6216 FPGA
Output
Tone
generator
Analogue
integrator
(to oscilloscope)
3500 2800 2550 14002100 1100 650 320 220 0
IN
10kHz1kHz
0 1000 2000 3000 4000 5000
2.00
1.80
1.60
1.40
1.20
1.00
0.80
0.60
0.40
0.20
0.00
Fitness
Generations
0
Generations
900
800
700
600
500
400
300
200
100
0
Hamming Distance
1000 2000 3000 4000 5000
In
Out
Out
In
In
Out
voltage
Average
Input period (ms)
2.5V
0.0V
5.0V
output
43.0 23.5
31.2
Case temperature
(Celsius)
F1 F2
1.41.21.00.80.60.40.20 1.6
configuration
PC
Desktop
XC6216 FPGA
Output
Tone
generator
Analogue
integrator
(to oscilloscope)
... Circuits can be evolved even with only a partial or high-level specification of the problem. Evolvable hardware is believed to have a lot of potential [191], for instance in adaptive hardware [78], or in fault-tolerant hardware [83], and it may allow to find novel or more efficient circuits than those obtained with traditional techniques [18,165,180]. ...
... This may lead to circuits that are more efficient than circuits that are designed (e.g. using less components). This was first demonstrated by Thompson that evolved a circuit capable of discerning between an input signal oscillating at two different frequencies [165,166]. This circuit was evolved so that its output was low when the input signal was oscillating at 1 kHz and high when the input signal was oscillating at 10 kHz. ...
... Output of the frequency discriminator at different generations for the 1 kHz and 10 kHz signals. After 3500 generations the input signal is correctly detected and the output is low with the 1 kHz signal, and high with the 10 kHz signal (from [165]). ...
Thesis
Full-text available
Bio-inspired electronic circuits have the potential to address some of the shortcomings of conventional electronic circuits, such as lack of applicability to ill-defined problems, of robustness, or of adaptivity to unexpectedly changing environments. Bio-inspired circuits are designed by taking inspiration from principles observed in biology. The evolution of biological organisms, their development from a fertilized egg, and their learning capabilities are three sources of bio-inspiration that can be used for this purpose. Until now bio-inspired electronics mostly focused on a single aspect of bio-inspiration: either evolution, development or learning. In this thesis we consider that electronic circuits should encompass all three aspects to fully benefit from bio-inspiration. These circuits capable of evolution, development and learning are called POEtic circuits (POE stands for phylogeny, ontogeny and epigenesis, that mean respectively evolution, development and learning). Conceptually these POEtic circuits, much like biological organisms, are multi-cellular circuits that evolve following the principles of selection and differential reproduction, they develop from a single cell and differentiate according to inter-cellular and environmental signals, and eventually they learn during their lifetime. These circuits may also dynamically reorganize their structure in order to cope with changes in the environment, or when they are expanded with new cells, sensors or actuators. In comparison to conventional circuits, POEtic circuits are created automatically using evolutionary principles, even if only a partial or high-level specification of the problem is known. Development provides a complex genotype to phenotype mapping, that may lead to fault-tolerance or adaptive development in order to cope with environmental changes. Finally learning allows these circuits to memorize past events or adapt their response over time to improve their behavior. This thesis deals with the evolutionary mechanisms required to evolve these POEtic circuits. We argue that in order to fully realize the potential of POEtic circuits a novel evolutionary system that takes into account their characteristics and that encompasses both a genetic encoding and a developmental system is required. Indeed, evolutionary algorithms commonly used to evolve electronic circuits do not exploit the complex dynamics of development which is seen in biological organisms. They generally use a direct genetic encoding with a one to one genotype to phenotype mapping. As a consequence the genetic string grows with the size of the circuits and this may limit the scalability of the evolutionary approach to larger circuits. Furthermore these encodings do not allow intercellular or environmental interactions during development, which could lead to adaptive development or fault-tolerant circuits. In this thesis we develop an evolutionary system suited for multi-cellular POEtic circuits. This evolutionary system is inspired by the mechanisms of gene expression and cellular differentiation seen in biological organisms. It attempts to provide better evolvability and scalability than direct genetic encodings, it allows cellular or environmental interactions during development, and it is computationally simple so that it can be efficiently implemented in hardware. It is furthermore generic, and it makes minimal assumptions on the circuits that are evolved: other than assuming they are multi-cellular, it only requires local communication between neighboring cells. We demonstrate the proposed evolutionary system by evolving multi-cellular circuits for a wide range of applications. The results that we obtain confirm the generality of our approach and its advantages in comparison to direct genetic encodings. The proposed evolutionary system is used to evolve structures of differentiated cells, and it shows better scalability to larger structures in terms of fitness than a direct genetic encoding. The dynamics of development within the evolutionary system can recover these structures in case of faults, even at high fault rates. The proposed evolutionary system is used to evolve multi-cellular circuits composed of spiking neurons to recognize patterns and to control the navigation with obstacle avoidance of a mobile robot, and in comparison it outperforms a direct genetic encoding. Finally it is used to evolve circuits capable of learning that control a mobile robot in a vision-based learning and navigation task. This last application demonstrates the three aspects of bio-inspiration of POEtic circuits in a single task: evolution, development and learning.
... This conceptual idea was named "evolution in materio" (EiM) by Miller and Downing (2002). Inspired by Thompson's seminal work on "intrinsic" hardware evolution (Thompson 1997), the principal idea is to use artificial evolution as a search method to find configurations that directly exploit the computational properties of complex materials. ...
Chapter
We overview Reservoir Computing (RC) with physical systems from an Unconventional Computing (UC) perspective. We discuss challenges present in both fields, including encoding and representation, or how to manipulate and read information; ways to search large and complex configuration spaces of physical systems; and what makes a “good” computing substrate.
... A second approach to constructing physical devices was initially explored in the early 1960s within the field of electrical engineering, originally for the purpose of recognizing printed letters [KL63] and has since been extended to many other fields [LAJ + 98, GCL + 03, EK10, ASKA18, MLP + 18], with sometimes very surprising results [Tho97]. In this approach, the designer specifies a mathematical objective function which, given a design, outputs a number representing how well the input design matches the desired specifications; the lower this number is, the better the design. ...
Preprint
In the photonic design problem, a scientist or engineer chooses the physical parameters of a device to best match some desired device behavior. Many instances of the photonic design problem can be naturally stated as a mathematical optimization problem that is computationally difficult to solve globally. Because of this, several heuristic methods have been developed to approximately solve such problems. These methods often produce very good designs, and, in many practical applications, easily outperform `traditional' designs that rely on human intuition. Yet, because these heuristic methods do not guarantee that the approximate solution found is globally optimal, the question remains of just how much better a designer might hope to do. This question is addressed by performance bounds or impossibility results, which determine a performance level that no design can achieve. We focus on algorithmic performance bounds, which involve substantial computation to determine. We illustrate a variety of both heuristic methods and performance bounds on two examples. In these examples (and many others not reported here) the performance bounds show that the heuristic designs are nearly optimal, and can considered globally optimal in practice.
Article
In the photonic design problem, a scientist or engineer chooses the physical parameters of a device to best match some desired device behavior. Many instances of the photonic design problem can be naturally stated as a mathematical optimization problem that is computationally difficult to solve globally. Because of this, several heuristic methods have been developed to approximately solve such problems. These methods often produce very good designs, and, in many practical applications, easily outperform 'traditional' designs that rely on human intuition. Yet, because these heuristic methods do not guarantee that the approximate solution found is globally optimal, the question remains of just how much better a designer might hope to do. This question is addressed by performance bounds or impossibility results, which determine a performance level that no design can achieve. We focus on algorithmic performance bounds, which involve substantial computation to determine. We illustrate a variety of both heuristic methods and performance bounds on two examples. In these examples (and many others not reported here) the performance bounds show that the heuristic designs are nearly optimal, and can be considered globally optimal in practice. This review serves to clearly set up the photonic design problem and unify existing approaches for calculating performance bounds, while also providing some natural generalizations and properties.
Article
Full-text available
The average builder in the USA provides a warrantee for 10 years, and the US Department for Energy calculates that US office buildings have an average lifespan of 73 years. No building is permanent, and all will face demolition at some point. When a building comes to the end of its safe and useful lifespan, there is no method for re-using the material in new buildings, instead, all constructions today require virgin material. This is a problem for sustainability because US cities, like most other global cities, require cyclical replacement of ageing buildings, and therefore perpetual resource extraction. This paper provides techniques for computationally arranging materials after the demolition and unmaking of architecture. Rather than downcycling concrete into low-value aggregate or melting float glass into opaque bottles methods are shown for this material to be indexed, re-machined and algorithmically arranged into new assemblies. These assemblies are conceived of as holding patterns; an indexed library of materials that are put into useful architectural arrangements, but ready to be disassembled towards some future use. These holding patterns are used as infill to the city rather than landfill beyond. Rather than building for sixty-year life spans, the project offers an imagination of eternal re-constructions that can learn from the carcass of past buildings. Based on rough estimates 2016 could be the first year where there exists more than one trillion tons of concrete on earth. More than the total weight of living trees on the planet (Crowther et al. 2015; USGS, 2018). This paper begins to develop new aptitudes for re-fitting misfit material rather than consuming evermore.
Article
Full-text available
Being able to identify the genetic variants responsible for specific bacterial phenotypes has been the goal of bacterial genetics since its inception and is fundamental to our current level of understanding of bacteria. This identification has been based primarily on painstaking experimentation, but the availability of large data sets of whole genomes with associated phenotype metadata promises to revolutionize this approach, not least for important clinical phenotypes that are not amenable to laboratory analysis. These models of phenotype-genotype association can in the future be used for rapid prediction of clinically important phenotypes such as antibiotic resistance and virulence by rapid-turnaround or point-of-care tests. However, despite much effort being put into adapting genome-wide association study (GWAS) approaches to cope with bacterium-specific problems, such as strong population structure and horizontal gene exchange, current approaches are not yet optimal. We describe a method that advances methodology for both association and generation of portable prediction models.
Article
Evolvable hardware (EH) architectures are capable of changing their configuration and behavior dynamically based on inputs from the environment. In this paper, we investigate the feasibility of using EH to prevent Hardware Trojan Horses (HTHs) from being inserted, activated or propagated in a digital electronic chip. HTHs are malicious hardware components that intend to leak secret information or cause malfunctioning at run-time in the chip in which they are integrated. We hypothesize that EH can detect internal circuit errors at run-time and reconfigure to a state in which the errors are no longer present. We implement a Virtual Reconfigurable Circuit (VRC) on a Field-Programmable Gate Array (FPGA) that autonomously and periodically reconfigures itself based on an Evolutionary Algorithm (EA). New VRC configurations are generated with an on-chip EA engine. We show that the presented approach is applicable in a scenario in which (1) the HTH-critical areas in the circuit are known in advance, and (2) the VRC is a purely combinatorial circuit, as opposed to the on-chip memory holding the golden reference, which requires one or more cycles to be read/written. We compare two different approaches for protecting the system against HTHs: Genetic Programming (GP) and Cartesian Genetic Programming (CGP). The paper reports on experiments on four benchmark circuits and gives an overview of both the limitations and the added value of the presented approaches.
Article
This paper presents the on‐chip evolution system of combinational logic circuits by a new hybrid algorithm known as improved genetic‐simulated annealing algorithm (IGASA). IGASA is based on the concept and principles of genetic algorithm (GA) and simulated annealing (SA). The main idea is to combine GA's global search ability with the advantage of SA's fast convergence to reach an optimal solution. In this paper, a new mutation operation and a repeated annealing (RA) process are introduced to overcome the premature convergence of the standard GA and the large number of searches in the later period. An elitist strategy to enhance convergence characteristic of the proposed algorithm is incorporated to better speed up the evolution process. In addition, the new algorithm is implemented in a new evolvable platform using Intel Cyclone V FPGA with an embedded ARM microprocessor. Comparison of the performance of the improved algorithm to the pure GASA algorithm and standard GA is presented on a number of case studies. The experiment results demonstrate that feasible circuits are always achieved by the IGASA algorithm unlike with other algorithms and the number of generations required is less.
Article
Full-text available
. Artificial evolution is discussed in the context of a successful experiment evolving a hardware configuration for a silicon chip (a Field Programmable Gate Array); the real chip was used to evaluate individual configurations on a tone-recognition task. The evolutionary pathway is analysed; it is shown that the population is genetically highly converged and travels far through genotype space. Species Adaptation Genetic Algorithms (SAGA) are appropriate for this type of evolution, and it is shown how an appropriate mutation rate was chosen. The role of junk on the genotype is discussed, and it is suggested that neutral networks (paths through genotype space via mutations which leave fitness unchanged) may be crucial to the effectiveness of evolution. 1 Introduction To evolve silicon hardware successfully one must understand the limitations and possibilities of hardware; but one must also understand the limitations and possibilities of artificial evolution. This paper focuse...
Article
Full-text available
: Artificial evolution can integrate fault tolerance considerations into the automatic design process, producing inherently fault-tolerant designs without explicit redundant parts. Population dynamics can give rise to some level of fault tolerance `for free.' Requirements for fault tolerance can also be incorporated into the fitness function. The practicalities of these methods are investigated, grounded in the study of a real-world evolved electronic control system for a robot. 1. INTRODUCTION This paper builds on earlier work [9] to show how artificial evolution --- such as a Genetic Algorithm (GA) --- can automatically design systems that are inherently tolerant to faults. In a harsh environment or a safety critical application, a system might be required to retain a certain level of ability even if a computer's memory becomes slightly corrupted, or a few transistors fail. Tolerance to semiconductor defects increases both yield and feasible chip size, and is a necessity for wafer-...
Book
1. A Neural Processor for Maze Solving.- 2 Resistive Fuses: Analog Hardware for Detecting Discontinuities in Early Vision.- 3 CMOS Integration of Herault-Jutten Cells for Separation of Sources.- 4 Circuit Models of Sensory Transduction in the Cochlea.- 5 Issues in Analog VLSI and MOS Techniques for Neural Computing.- 6 Design and Fabrication of VLSI Components for a General Purpose Analog Neural Computer.- 7 A Chip that Focuses an Image on Itself.- 8 A Foveated Retina-Like Sensor Using CCD Technology.- 9 Cooperative Stereo Matching Using Static and Dynamic Image Features.- 10 Adaptive Retina.
Article
This is the thoroughly revised and updated second edition of the hugely successful The Art of Electronics. Widely accepted as the single authoritative text and reference on electronic circuit design, both analog and digital, the original edition sold over 125,000 copies worldwide and was translated into eight languages. The book revolutionized the teaching of electronics by emphasizing the methods actually used by citcuit designers - a combination of some basic laws, rules to thumb, and a large nonmathematical treatment that encourages circuit values and performance. The new Art of Electronics retains the feeling of informality and easy access that helped make the first edition so successful and popular. It is an ideal first textbook on electronics for scientists and engineers and an indispensable reference for anyone, professional or amateur, who works with electronic circuits. The best self-teaching book and reference book in electronics Simply indispensable, packed with essential information for all scientists and engineers who build electronic circuits Totally rewritten chapters on microcomputers and microprocessors The first edition of this book has sold over 100,000 copies in seven years, it has a market in virtually all research centres where electronics is important
Article
Excerpt If we are asked where the transition from nonlife to life occurred, we must first refer to the conceptual level of DNA, of RNA, or of any of their self-replicating precursors. If we take any class of chemical compounds that are composed of a finite number of monomeric types, which we may call symbols, and if these symbols are lined up in linear sequences, thereby representing messages that can be “read” through specific interactions (i.e., complementary base pairing), then our class of compounds embodies a new property that goes beyond chemistry: This property is information. Such a quality can no longer be called typical for chemistry; it is characteristic of biology. The property of self-reproduction—unique to DNA, to RNA, or to any analogous precursor polymer—has two consequences. (1) The information stored in the sequence of symbols is conserved in replication; this provides a genetic memory that resists the destruction to...
Article
Evolution of RNA secondary structure is studied using simulation techniques and statistical analysis of fitness landscapes. The transition from RNA sequence to RNA secondary structure leads to fitness landscapes that have local variations in their "ruggedness." Evolution exploits these variations. In stable environments it moves the quasispecies toward relatively "flat" peaks, where not only the master sequence but also its mutants have a high fitness. In a rapidly changing environment, the situation is reversed; evolution moves the quasispecies to a region where the correlation between secondary structures of "neighboring" RNA sequences is relatively low. In selection for simple secondary structures the movement toward flat peaks leads to pattern generation in the RNA sequences. Patterns are generated at the level of polynucleotide frequencies and the distribution of purines and pyrimidines. The patterns increase the modularity of the sequence. They thereby prevent the formation of alternative secondary structures after mutations. The movement of the quasispecies toward relatively rugged parts of the landscape results in pattern generation at the level of the RNA secondary structure. The base-pairing frequency of the sequences increases. The patterns that are generated in the RNA sequences and the RNA secondary structures are not directly selected for and can be regarded as a side effect of the evolutionary dynamics of the system.