No abstract

Targeted at high-energy physics research applications, our special-purpose analog neural processor can classify up t o 70 dimensional vectors within 50 nanoseconds. The decisionmaking process of the implemented feedforward neural network enables this type of computation to tolerate weight discretization, synapse nonlinearity, noise, and other nonideal effects. Although our prototype does not take advantage of advanced CMOS technology, and was fabricated using a 2.5-pm CMOS process, it performs 6 billion multiplications per second, with only 2W dissipation, and has as high as 1.5 Gbyte/s equivalent bandwidth.lthough neural networks offer exceptionally powerful parallel computation performance, most current applications focus on exploiting their leaming capabilities. The ability of neurai networks to learn from examples has given rise to several quite successful experiments. Those involving handwritten character recognition, speech recognition, and similar challenges in which biological systems overshadow artificial intelligence come to mind.Still, the benefits of unique parallel processing that neural networks afford warrant our attention as well. With fully parallel neural hardware, processing time is independent of the data-size the network must process. Only a few computing steps require serial processing, making computation time extremely short. The inner product computation involved does present one major challenge for realizing the hardware of neural nets. Therefore, if an application does not demand high precision, the compact, high-speed analog approach provides great advantages.Analog techniques let us create single-chip architectures of complex neural networks, featuring low cost and low power dissipation. Such hardware, offering processing times as low as several microseconds for as large as 128 dimensional input vectors, is already commercially available.' Still, solving for application domains that demand tens of nanoseconds of processing delayzt3 for similarly large input vectors is almost impossible. The new architecture this article describes provides precisely this sort of highcomputing performance, offering one solution for a number of demanding applications. Nuclear research applicationsTo help understand the behavior of fundamental particles and forces, Hamburg's High Energy Physics (HEP) Institute Deutsches Elektronen Synchrotron (DESY) operates two large detectors installed within its hadron-electron ring accelerator (HEM). They are called H1 and Zeus. The two detectors contain different components, each specialized for detecting track, momentum, or energy of particles coming from the interaction region, where electrons and protons collide.These detectors provide tremendous amounts of information through 200,000 analog channels, sampled at a rate of 10 million times a second and producing 10l6 bytes of data per second. The resulting data flow, which requires real-time processing, imposes a great challenge for the dataacquisition system, far exceeding the capabilities even of available superco...

Articles you may be interested inHigh-percentage success method for preparing and pre-evaluating tungsten tips for atomic-resolution scanning tunneling microscopyWe show that a certain type of artificial neural systems or neural networks, more specifically the backpropagation network (BPN), can be an efficient tool in fast, approximate pre-evaluation of spectroscopic ellipsometric (SE) measurements. The BPN is a multilayer, feedforward network which can perform nontrivial mapping functions. We demonstrate the method on separation by implantation of oxygen (SIMOX) structure and ion implantation caused damage depth profile evaluation. The results are compared with others from independent measurements.

Abstmcf -A CMOS neural network integrated circuit is discussed, which was designed for very high speed applications. This full-custom, mixed analog-digital chip implements a fully connected feedforward neural network with 70 inputs, 6 hidden layer neurons and one output neuron. The neurons perform inner product operation and have sigmoid-like activation function. The 70 network inputs and the neural signal processing are analog, the synaptic weights are digitally programmable with 5 bit (4 bits + sign) precision.The synaptic weights are stored on onchip static R A M cells. The combination of analog and digital techniques results Unique computing power with ease of use. Progrrunming can easily be performed with the help of a spreadsheet or other suitable Interface program from a PC. The resolution of the input signals is mainly determined by the signal to noise ratio which lies typically between 8-12 bits. Therefore the equivalent Input bandwidth can be as high as 28-42 GbiWsecond. The system is designed for very high speed vector classification and the feasibility of a Single chlp neural network photon trigger for nudear research is shown. Because of the fully parallel architecture and the fast analog signal processing the network achieves unique computing performance and clasdfles up to 70 dimensional vectors within 20 nanoseconds, performing 20 billion (2*1010) multiply-and-add operations per second. The circuit occupies l0r9mm2 silicon area with 1.5p.m CMOS process and Bissip8tes only 1W at SV supply.

A CMOS neural network IC is discussed, which was designed for very high speed applications. The parallel architecture, analog computing and digital weight storage provides unprecedented computing speed combined with ease of use. The circuit classifies up to 70 dimensional vectors within 20 nanoseconds, performing 20 billion (2*1 Ole) multiply-and-add operations per second, and has as high as 28-42 Gbitslsecond equivalent input bandwidth with less than 1W dissipation. The synaptic weights can be directly downloaded from a host computer to the on onchip SRAM. The full-custom, analog-digital chip implements a fully connected feedforward neural network with 70 inputs, 6 hidden layer neurons and one output neuron. A unique solution, a single chip neural network photon trigger for high-energy physics research is provided.