Ádám Rák

Macromodeling of the Memristor in SPICE

In this paper, we present a new simulation program with integrated circuit emphasis macromodel of the recently physically implemented memristor. This macromodel could be a powerful tool for electrical engineers to design and experiment new circuits with memristors. Our simulation results show similar behavior to the already published measurements of the physical implementation. Our approach provides a solution for the modeling of boundary conditions following exactly the published mathematical model of HP Labs. The functionality of our macromodel is demonstrated with computer simulations. The source code of our macromodel is provided.

GPU powered CNN simulator (SIMCNN) with graphical flow based programmability

In this paper, we introduce an innovative CNN algorithm development environment that significantly assists algorithmic design. The introduced graphical user interface uses Matlab Simulink with UMF-like program description, where direct functionality accompanies better accessability. The new generation of graphical cards incorporate many general purpose graphics processing units, giving the power of parallel computing to a simple PC environment cheaply. Therefore, analysis of CNN dynamics become more feasible with a common hardware setup. Our measurements demonstrate the efficiency of the realized system. In the case of simpler algorithms, real-time execution is also possible.

GPU boosted CNN simulator library for graphical flow-based programmability

A graphical environment for CNN algorithm development is presented. The new generation of graphical cards with many general purpose processing units introduces the massively parallel computing into PC environment. Universal Machine on Flows- (UMF) like notation, highlighting image flows and operations, is a useful tool to describe image processing algorithms. This documentation step can be turned into modeling using our framework backed with MATLAB Simulink and the power of a video card. This latter relatively cheap extension enables a convenient and fast analysis of CNN dynamics and complex algorithms. Comparison with other PC solutions is also presented. For single template execution, our approach yields run times 40x faster than that of the widely used Candy simulator. In the case of simpler algorithms, real-time execution is also possible.

Cellular neural networks with memristive cell devices

In this paper, we present simulation measurements of a memristor crossbar device. We designed a PCB memristor package and the appropriate measurement board. Technical details of these circuits are presented. Cellular like topology of this crossbar device can provide high density and local connectivity. We gave a formula to evaluate the direction of the change of the states of the memristor array in case of a given voltage input. Our simulation results show that a memristor crossbar can be a trainable weight-matrix of a fully connected neural network if the memristors have ohmic non-linearity.

Chua's Circuit Topology Evolution Using Genetic Algorithm

In this paper, a novel approach of using genetic algorithm towards the realization of Chuas circuit is presented. The paper demonstrates the use of genetic algorithm to evolve Chuas circuit from a set of given passive and active components. More than a dozen of the evolved designs are demonstrated to work satisfactorily in simulation. The evolved designs are found to be competitive in a human environment and are also demonstrated to work in a laboratory set up.

CPU-GPU hybrid compiling for general purpose: Case studies

In this demonstration, a new approach to GPU utilization has been demonstrated. The compilation of the standard C or C++ source code with our GCC based compiler results hybrid executable which uses GPU as well. Host CPU code is generated automatically. Case studies are presented.

Energy calculation of MMFF94 force field on GPU (Special session on array computing systems and applications)

One important field of computer simulations is material science, the examination of the structure and interaction of molecules, for which various models and algorithms have been developed. One of these is the MMFF94 force field, which describes potential energies. Gradually modifying the coordinates of the atoms in the molecule and calculating the energy, states with favorable energies can be found. These could describe the real structure. In this work the implementation of the energy calculation of MMFF94 force field on GPU is described and analyzed.

Independent component analysis by memristor based neural networks

In this paper, we propose a memristor based circuit for independent component analysis. The structure of our circuit is described in detail. The simulation results show that our memristor based circuit structure can learn blind source separation in an unsupervised way.

Emulation of analog memristors using low yield digital switching memristors

In this paper, we propose circuits for emulating analog purpose memristors (APM) purely composed from digital switching memristors or digital purpose memristors (DPM). Our design is also robust against memristor defects therefore this can be applied even in low yield technology where defects occur at high probability. Our simulations show that the switching noise of the circuits decreases with the number of memristors and the production yield of the circuit is no lower than the production yield of the digital switching memristors.