András Horváth

Spin torque oscillator models for applications in associative memories

We present physics-based models for both individual and coupled spin torque nano oscillators (STNOs). Such STNOs may become as building blocks for CNN-like dynamic computing architectures. We discuss a hierarchy of models, extending from micromagnetic models, which include the detailed geometry and physics, to compact models, which are based on parameters extracted from the underlying physical description. These simulations also include coupling between individual STNOs, both via spin waves and via electrical interconnects. Using this modeling approach, we demonstrate frequency entrainment and phase synchronization between STOs in the array, which enable computing functions.

Correlated confocal and super-resolution imaging by VividSTORM

Single-molecule localization microscopy (SMLM) is rapidly gaining popularity in the life sciences as an efficient approach to visualize molecular distribution with nanoscale precision. However, it has been challenging to obtain and analyze such data within a cellular context in tissue preparations. Here we describe a 5-d tissue processing and immunostaining procedure that is optimized for SMLM, and we provide example applications to fixed mouse brain, heart and kidney tissues. We then describe how to perform correlated confocal and 3D-superresolution imaging on these sections, which allows the visualization of nanoscale protein localization within labeled subcellular compartments of identified target cells in a few minutes. Finally, we describe the use of VividSTORM (http://katonalab.hu/index.php/vividstorm), an open-source software for correlated confocal and SMLM image analysis, which facilitates the measurement of molecular abundance, clustering, internalization, surface density and intermolecular distances in a cell-specific and subcellular compartment–restricted manner. The protocol requires only basic skills in tissue staining and microscopy.

An Associative Memory with oscillatory CNN arrays using spin torque oscillator cells and spin-wave interactions architecture and End-to-end Simulator

An Associative Memory is built by three consecutive components: (1) a CMOS preprocessing unit generating input feature vectors from picture inputs, (2) an AM cluster generating signature outputs composed of spintronic oscillator (STO) cells and local spin-wave interactions, as an oscillatory CNN (O-CNN) array unit, applied several times arranged in space, and (3) a classification unit (CMOS). The end to end design of the preprocessing unit, the interacting O-CNN arrays, and the classification unit is embedded in a learning and optimization procedure where the geometric distances between the STOs in the O-CNN arrays play a crucial role. The O-CNN array has an input vector as a 1D array of oscillator frequencies, and the synchronized O-CNN array codes the output as the phases of the output 1D array. The typical O-CNN array has 1-3 rows of STOs. Simplified STO and interaction macro models are used. A typical example is shown using an End-to-end Simulator.

Synchronization in cellular spin torque oscillator arrays

Spin torque nanodevices could provide a platform for computation beyond Moores law. The network of spin oscillators can have only local, cellular interconnections because of the underlying physics: the interaction between the oscillators happens through the magnetic field. In this paper we describe the dynamics of weakly coupled spin-torque oscillator networks and how the dynamics of these cellular arrays can be used for problem solving. We will describe how the phase shift in a synchronized array can be calculated between the elements and we will also show a simple example how the dynamics of a cellular array can be used to solve simple tasks.

Impact of steep-slope transistors on non-von Neumann architectures: CNN case study

A Cellular Neural Network (CNN) is a highly-parallel, analog processor that can significantly outperform von Neumann architectures for certain classes of problems. Here, we show how emerging, beyond-CMOS devices could help to further enhance the capabilities of CNNs, particularly for solving problems with non-binary outputs. We show how CNNs based on devices such as graphene transistors - with multiple steep current growth regions separated by negative differential resistance (NDR) in their I-V characteristics - could be used to recognize multiple patterns simultaneously. (This would require multiple steps given a conventional, binary CNN.) Also, we demonstrate how tunneling field effect transistors (TFETs) can be used to form circuits capable of performing similar tasks. With this approach, more “exotic” device I-V characteristics are not required - which should be an asset when considering issues such as cell-to-cell mismatch, etc. As a case study, we present a CNN-cell design that employs TFET-based circuitry to realize ternary outputs. We then illustrate how this hardware could be employed to efficiently solve a tactile sensing problem. The total number of computation steps as well as the required hardware could be reduced significantly when compared to an approach based on a conventional CNN.

Dynamic coupling of spin torque oscillators for associative memories

Spin torque nanodevices could be possible candidates for future architectures, they small size and low power consumption makes them a possible option for beyond Moores law computation. Coupling of individual oscillators can be done by many different methods (e.g.: spin-wave coupling or electrical coupling). These couplings are usually static and wired for a specific array, resulting a specific functionality and it can not be changed dynamically. Coupling a large number of oscillators will require a large number of coupling elements for programmable coupling. In this paper we will show who spin torque oscillators can be coupled by common frequencies via a common coupling signal. This way a large number of oscillators can be coupled simultaneously and their coupling can be altered dynamically.

Architectural impacts of emerging transistors

At present there is much effort to determine if emerging information processing devices could have a positive impact on the performance of non-Boolean/non-von Neumann computer architectures. We explore this topic here by specifically considering how emerging transistor technologies might impact cellular neural networks (CNNs). For CNNs, prior work suggests that new transistor structures could be employed to better facilitate non-binary outputs - that in turn reduce the number of template/programming operations as well as the hardware paths needed to solve a given problem. Here, we present analysis that considers how the above approach could impact the performance/energy of the cells that comprise the CNN. We also consider how characteristics of other emerging transistor technologies could positively impact CNNs - particularly with respect to reduced program complexity.

TFET-based Operational Transconductance Amplifier Design for CNN Systems

A Cellular Neural Network (CNN) is a powerful processor that can significantly improve the performance of spatio-temporal applications such as pattern recognition, image processing, motion detection, when compared to the more traditional von Neumann architecture. In this paper, we show how tunneling field effect transistors (TFETs) can be utilized to enhance the performance of CNNs. Specifically, power consumption of TFET-based CNNs can be significantly lower when compared to MOSFET-based CNNs due to improved voltage controlled current sources (VCCSs) - an important component in CNN systems. We demonstrate that CNNs can benefit from low power conventional linear VCCSs implemented via TFETs. We also show that TFETs can be useful to realize non-linear VCCSs, which are either not possible or exhibit degraded performance when implemented via CMOS. Such non-linear VCCSs help to improve the performance of certain CNN operations (e.g., global maximum/minimum). We provide two case studies - image contrast enhancement and maximum row selection - that illustrate the benefits of non-linear VCCSs (e.g., reduced computation time, energy dissipation, etc.) when compared to CMOS-based approaches.

Fast, parallel implementation of particle filtering on the GPU architecture

In this paper, we introduce a modified cellular particle filter (CPF) which we mapped on a graphics processing unit (GPU) architecture. We developed this filter adaptation using a state-of-the art CPF technique. Mapping this filter realization on a highly parallel architecture entailed a shift in the logical representation of the particles. In this process, the original two-dimensional organization is reordered as a one-dimensional ring topology. We proposed a proof-of-concept measurement on two models with an NVIDIA Fermi architecture GPU. This design achieved a 411- μ s kernel time per state and a 77-ms global running time for all states for 16,384 particles with a 256 neighbourhood size on a sequence of 24 states for a bearing-only tracking model. For a commonly used benchmark model at the same configuration, we achieved a 266- μ s kernel time per state and a 124-ms global running time for all 100 states. Kernel time includes random number generation on the GPU with curand. These results attest to the effective and fast use of the particle filter in high-dimensional, real-time applications.

Fast computation of particle filters on processor arrays

We have developed a new variant of the particle filter algorithm for estimating a signal from noisy observations. It suits ideally implementation on a cellular processor array. The error of the new algorithm is essentially the same as that of the old one but it runs much faster, especially when there is a large number of particles to be simulated.