摘要:
The processing element is useful in solving problems from the class of temporal signal processing problems and is modeled closely on the sub-cellular biology and electrophysiology of neurons having chemical synapses.
摘要:
Une cellule synaptique semi-conductrice, à domaine de charge (10), comprend un condensateur (15) couplé entre une ligne d'entrée (11) et un noeud intermédiaire (17). Une transition descendante de tension, appliquée à la ligne d'entrée (11), produit un transfert de charge d'une ligne de sommation (S1) vers le noeud intermédiaire (17) par l'intermédiaire d'un premier dispositif (14) présentant un seuil programmable. Un second dispositif (13) transfère la charge du noeud intermédiaire (17) vers une autre ligne de sommation (S2) en réponse à la prochaine transition descendante de tension de la ligne d'entrée. La quantité de charge transférée est proportionnelle à l'amplitude du signal d'entrée pulsé, au seuil de tension programmé et à la valeur de capacitance.
摘要:
The present invention relates to a product-sum operation circuit element and a circuit for addition by weighting a number of signals input in one neuron circuit (N₁, N₂, ... N n ) in a neural network, and can provide an adaptive-learning neuron circuit (N₁, N₂, ... N n ) for changing an interval of output pulses by learning by connecting a simple pulse generating circuit consisting of at least capacitance, resistance, unijunction transistor UJT and MISFET. A product-sum operation circuit element according to the present invention, comprises an insulator substrate (35), a single crystal semiconductor thin film (36, 37) having a p-n-p or n-p-n structure in a lateral direction formed in the shape of stripes on the insulator substrate (35), a ferroelectric thin film (38) deposited thereon for covering at least the semiconductor stripe structure (36, 37), and a stripe-like gate electrode 39 consisting of a metal or a polycrystalline semiconductor formed thereon for intersecting said semiconductor stripes at a right angle or suitable angle.
摘要:
Procédé de connexion interne pour réseaux de neurones reliant des couches de neurones successives. Les états de sortie des neurones d'une nouvelle couche sont représentés par des fonctions obtenues suivant une première étape (21) en effectuant une sommation pondérée de fonctions représentant chacune l'état de sortie d'un neurone de la couche précédente, suivant une deuxième étape (22) en appliquant une fonction saturante au résultat de la sommation et suivant une troisième étape (23) en appliquant une fonction de distribution au résultat de la fonction saturante. Application : traitement et analyse d'image, traitement du signal.
摘要:
An electronic image processing system uses data provided by one or more sensors to perform cooperative computations and improve image recognition performance. A smoothing resistive network, which may comprise an integrated circuit chip, has switching elements connected to each node. The system uses a first sensory output comprising primitives, such as discontinuities or object boundaries, detected by at least a first sensor to define a region for smoothing of a second sensory output comprising at least a second, distinct output of the first sensor or a distinct output of at least a second sensor. A bit pattern for controlling the switches is generated from the detected image discontinuities in the first sensory output. The second sensory output is applied to the resistive network for data smoothing. The switches turned off by the data from the first sensory output define regional boundaries for smoothing of the data provided by the second sensory output. Smoothing operations based on this sensor fusion can proceed without spreading object characteristics beyond the object boundaries.
摘要:
A semiconductor device comprising on a substrate a first semiconductor region of one conductive type, first source and and drain regions of the opposite conductive type formed in the above regions, a first gate electrode formed in a region separating said source and drain regions, the first gate electrode being electrically floated through an insulating film, and at least two second gate electrodes connected to said gate electrode by capacitive coupling, wherein an inverted layer is formed under said first gate electrode and said first source and drain regions are electrically connected together only when a predetermined threshold value is exceeded by an absolute value of a value obtained by linearly summing up the weighed voltages applied to the second gate electrodes.
摘要:
The use of neural networks has been employed to adjust processing during the fabrication of articles. For example, in the production of photolithographic masks by electron beam irradiation of a mask blank in a desired pattern, electrons scattered from the mask substrate cause distortion of the pattern. Adjustment for such scattering is possible during the manufacturing process by employing an adjustment function determined by a neural network whose parameters are established relative to a prototypical mask pattern.
摘要:
A neural coupling is composed of a pair of primary and secondary members (13,14) for coupling a precedent node (10) to a subsequent node (11) to synaptically transmit a signal through a neural network. A primary member is receptive of a signal from a precedent node for generating an inductive field according to the received signal. A secondary member is spatially coupled to the primary member through the inductive field for producing to a subsequent node a corresponding signal in response to the inductive field so as to effect synaptic transmission of the signal through the neural network.
摘要:
A neural network utilizing the threshold characteristics of a semiconductor device as the various memory elements of the network. Each memory element comprises a complementary pair of MOSFETs in which the threshold voltage is adjusted as a function of the input voltage to the element. The network is able to learn by example using a local learning algorithm. The network includes a series of output amplifiers in which the output is provided by the sum of the outputs of a series of learning elements coupled to the amplifier. The output of each learning element is the difference between the input signal to each learning element and an individual learning threshold at each input. The learning is accomplished by charge trapping in the insulator of each individual input MOSFET pair. The thresholds of each transistor automatically adjust to both the input and output voltages to learn the desired state. After input patterns have been learned by the network, the learning function is set to zero so that the thresholds remain constant and the network will come to an equilibrium state under the influence of a test input pattern thereby providing, as an output, the learned pattern most closely resembling the test input pattern.
摘要:
An analog neural network composed of an array of capacitors (C1j, C2j...Cnj) for storing weighted electric charges. Electric charges, or voltages, on the capacitors control the impedance (resistance) values of a corresponding plurality of MOSFETs (28, 29...30) which selectively couple input signals to one input of a summing amplifier (31). A plurality of semiconductor gating elements (e.g. MOSFETs) (25, 26..27) selectively couple to the capacitor's weighted analog voltage values received sequentially over an input line (18). The weighted voltages on the input line are periodically applied to the proper capacitors in the neural network via the gating elements so as to refresh the weighted electric charges on the capacitors, and at a multiplex rate that maintains the voltages on the capacitors within acceptable tolerance levels.