Part of the  

Chip Design Magazine

  Network

About  |  Contact

Posts Tagged ‘Perceptron’

Cadence Puts a Neural Network in a DSP

Monday, May 1st, 2017

Gabe Moretti, Senior Editor

Cadence Design Systems, Inc. today unveiled the Cadence Tensilica Vision C5 DSP, the industry’s first standalone, self-contained neural network DSP IP core optimized for vision, radar/lidar and fused-sensor applications with high-availability neural network computational needs. Targeted for the automotive, surveillance, drone and mobile/wearable markets, the Vision C5 DSP offers 1TMAC/sec computational capacity to run all neural network computational tasks.

What is a Neural Network?

The neural network technology mimics our present understanding of how a human brain works.  Figure 1 shown a depiction of a neuron, the component of a neural network.

Figure 1: A biological neuron

The neuron takes inputs from the dendrites, process them and send the output through the axon to be distributed by the boutons.  The “signals” are propagated and operated upon throughout the network, which operates as a pattern recognition machine.  The digital computational equivalent is shown in Figure 2.  It is important to understand that a neuron network is a directed flowgraph, so that the output of a node impact all of the following connected nodes in the graph, and that the signals are unidirectional.

FFigure 2; A Perceptron (Artificial Neuron)

The machine equivalent of the nucleus of the neuron, called a perceptron, has one or more inputs, computational capabilities, and an output that can be distributed to one or more nodes in the flowgraph. A typical perceptron has many inputs and these inputs are all individually weighted. The perceptron weights can either amplify or de-amplify the original input signal. For example, if the input is 1 and the input’s weight is 0.2 the input will be decreased to 0.2. These weighted signals are then added together and passed into the activation function. The activation function is used to convert the input into a more useful output. There are many different types of activation function.  Neural networks are very efficient for machine vision applications.