Towards Intelligently Designed Evolvable Processors

Abstract

Evolution-in-Materio is a computational paradigm in which an algorithm reconfigures a material's properties to achieve a specific computational function. This paper addresses the question of how successful and well performing Evolution-in-Materio processors can be designed through the selection of nanomaterials and an evolutionary algorithm for a target application. A physical model of a nanomaterial network is developed which allows for both randomness, and the possibility of Ohmic and non-Ohmic conduction, that are characteristic of such materials. These differing networks are then exploited by differential evolution, which optimises several configuration parameters (e.g., configuration voltages, weights, etc.), to solve different classification problems. We show that ideal nanomaterial choice depends upon problem complexity, with more complex problems being favoured by complex voltage dependence of conductivity and vice versa. Furthermore, we highlight how intrinsic nanomaterial electrical properties can be exploited by differing configuration parameters, clarifying the role and limitations of these techniques. These findings provide guidance for the rational design of nanomaterials and algorithms for future Evolution-in-Materio processors.

In-Materio Extreme Learning Machines

“…Such a DE algorithm can be combined with a material simulation (developed in [21]) to allow for significantly faster testing and analysis of EiM processors than physical manufacturing and experimentations would allow. Full details are available elsewhere [37,10], but briefly, the DE algorithm uses the greedy criterion that involves evaluating the fitness of each member of a generation's population, with those members of the population with better fitness being more likely to proceed to the next generation.…”

“…The output voltages (i.e., material processor output states) are regressed to produce an output layer which predicts the class (ŷ) of the processed data instances. Input weights (lr) can be used to scale the input data voltages, and configurable voltage stimuli (Vc) can manipulate the processor's behaviour [21].…”

“…voltages have been shown to both introduce a bias but also alter the decision boundary of an EiM classifier [21]; therefore, these effect how the materials IV characteristic is exploited which could be analogous to altering the material's 'activation function'. Input weights l r ∈ [−1, 1], are used to scale the input voltages V in r applied at the data driven input electrodes r due to a corresponding input attribute a r , such that:…”

“…If this relative computational simplicity is paired with low power materials, then in-Materio processors present a possible candidate for edge case computing [30]. Recent analysis of EiM systems has established fundamental good practices [21] and enhancements [20] to the EiM paradigm. However, even with these improvements, methods to scale in-Materio processing systems to larger and more complex datasets remains lacking.…”

“…Networks of common electronic devices (resistors, diodes, etc.) can provide interesting tunable dynamics [22] and can be realised physically or investigated using reliable and fast SPICE (Simulation Program with Integrated Circuit Emphasis) simulations [21,20].…”