Search

CN-121981182-A - Semiconductor device and electronic apparatus

CN121981182ACN 121981182 ACN121981182 ACN 121981182ACN-121981182-A

Abstract

The present disclosure relates to a semiconductor device and an electronic apparatus. The neuron circuit is capable of switching between two functions, a function as an input neuron circuit and a function as a hidden neuron circuit. The error circuit is capable of switching between a function as a hidden error circuit and a function as an output neuron circuit. The switching circuit is configured to be able to change the connection between the neuron circuit, the synaptic circuit and the error circuit. The burst circuit includes an analog memory storing data corresponding to a connection strength between the input neuron circuit and the hidden neuron circuit or between the hidden neuron circuit and the output neuron circuit, a write circuit changing data in the analog memory, and a weighting circuit weighting an input signal according to the data of the analog memory and outputting the weighted output signal. The analog memory includes a transistor including an oxide semiconductor having an extremely low off-state current.

Inventors

  • Kategawa

Assignees

  • 株式会社半导体能源研究所

Dates

Publication Date
20260505
Application Date
20161018
Priority Date
20151023

Claims (15)

  1. 1. A semiconductor device has a first module and a second module, wherein, The first module has a first circuit and a second circuit, The first circuit has a selection circuit that, The first circuit has a first function of outputting an analog first signal input to the first circuit to the second circuit via the selection circuit, and a second function of generating an analog second signal from the first signal by converting a current to a voltage and outputting the second signal to the second module via the selection circuit, The second circuit has an analog memory, a first multiplication circuit and a second multiplication circuit, The analog memory has a function of storing data corresponding to the connection strength, The first multiplication circuit has a function of generating a third signal by weighting the first signal input to the second circuit according to the data, The second multiplication circuit has a function of updating the data stored in the analog memory.
  2. 2. A semiconductor device has a first module and a second module, wherein, The first module has a first circuit, a second circuit and a third circuit, The first circuit has a selection circuit that, The first circuit has a first function of outputting an analog first signal input to the first circuit to the second circuit via the selection circuit, and a second function of generating an analog second signal from the first signal by converting a current to a voltage and outputting the second signal to the second module via the selection circuit, The second circuit has an analog memory, a first multiplication circuit and a second multiplication circuit, The analog memory has a function of storing data corresponding to the connection strength, The first multiplication circuit has a function of generating a third signal by weighting the first signal input to the second circuit according to the data, The second multiplication circuit has a function of updating the data stored in the analog memory using a fourth signal and the first signal input to the second circuit, The third circuit has a first function of generating a fifth signal from the third signal by converting a current into a voltage, and a second function of generating the fourth signal from a difference between the fifth signal and a sixth signal.
  3. 3. A semiconductor device has a first circuit and a second circuit, The first circuit has a selection circuit that, The first circuit has a first function of outputting an analog first signal input to the first circuit to the second circuit via the selection circuit, and a second function of generating an analog second signal from the first signal by converting a current to a voltage and outputting the second signal to the second circuit via the selection circuit, The second circuit has an analog memory, a first multiplication circuit and a second multiplication circuit, The analog memory has a function of storing data corresponding to the connection strength, The first multiplication circuit has a function of generating a third signal by weighting the first signal or the second signal input to the second circuit according to the data, The second multiplication circuit has a function of updating the data stored in the analog memory.
  4. 4. A semiconductor device has a first circuit, a second circuit, and a third circuit, The first circuit has a selection circuit that, The first circuit has a first function of outputting an analog first signal input to the first circuit to the second circuit via the selection circuit, and a second function of generating an analog second signal from the first signal by converting a current to a voltage and outputting the second signal to the second circuit via the selection circuit, The second circuit has an analog memory, a first multiplication circuit and a second multiplication circuit, The analog memory has a function of storing data corresponding to the connection strength, The first multiplication circuit has a function of generating a third signal by weighting the first signal or the second signal input to the second circuit according to the data, The second multiplication circuit has a function of updating the data stored in the analog memory using a fourth signal and the first signal input to the second circuit, The third circuit has a first function of generating a fifth signal from the third signal by converting a current into a voltage, and a second function of generating the fourth signal from a difference between the fifth signal and a sixth signal.
  5. 5. The semiconductor device according to any one of claims 1 to 4, wherein, The first circuit is a neuron circuit and the second circuit is a synaptic circuit.
  6. 6. A semiconductor device has a first module and a second module, wherein, The first module has a neuron circuit and a synapse circuit, The neuron circuit has a selection circuit which, The neuron circuit has a first function of outputting an analog first signal input to the neuron circuit to the synapse circuit via the selection circuit, and a second function of generating an analog second signal from the first signal by converting a current to a voltage and outputting the second signal to the second module via the selection circuit, The synaptic circuit has an analog memory.
  7. 7. A semiconductor device has a neuron circuit and a synaptic circuit, The neuron circuit has a selection circuit which, The neuron circuit has a first function of outputting an analog first signal input to the neuron circuit to the synapse circuit via the selection circuit, and a second function of generating an analog second signal from the first signal by converting a current into a voltage and outputting the second signal to the synapse circuit via the selection circuit, The synaptic circuit has an analog memory.
  8. 8. A semiconductor device has a neuron circuit and a synaptic circuit, wherein, The neuron circuit has a function of inputting the neuron circuit and a function of hiding the neuron circuit, The neuron circuit has a selection circuit which, When the neuron circuit is switched to the function of the input neuron circuit by the selection circuit, the neuron circuit outputs an analog first signal input to the neuron circuit to the synapse circuit via the selection circuit, When the neuron circuit is switched to the function of the hidden neuron circuit by the selection circuit, the neuron circuit generates an analog second signal from the first signal by converting a current to a voltage, and outputs the second signal to the synapse circuit via the selection circuit, The synaptic circuit has an analog memory, a first multiplication circuit and a second multiplication circuit, The analog memory has a function of storing data corresponding to the connection strength, The first multiplication circuit has a function of weighting the first signal according to the data, The second multiplication circuit has a function of updating the data stored in the analog memory.
  9. 9. A semiconductor device has a neuron circuit, a synaptic circuit, and an error circuit, wherein, The neuron circuit has a function of inputting the neuron circuit and a function of hiding the neuron circuit, The neuron circuit has a selection circuit which, When the neuron circuit is switched to the function of the input neuron circuit by the selection circuit, the neuron circuit outputs an analog first signal input to the neuron circuit to the synapse circuit via the selection circuit, When the neuron circuit is switched to the function of the hidden neuron circuit by the selection circuit, the neuron circuit generates an analog second signal from the first signal by converting a current to a voltage, and outputs the second signal to the synapse circuit via the selection circuit, The synaptic circuit has an analog memory, a first multiplication circuit and a second multiplication circuit, The analog memory has a function of storing data corresponding to the connection strength, The first multiplication circuit has a function of generating a second signal by weighting the first signal according to the data, The second multiplication circuit has a function of updating the data stored in the analog memory using an error signal and the first signal, The error circuit has a function of outputting a neuron circuit and a function of hiding the error circuit.
  10. 10. The semiconductor device according to claim 9, wherein, The error circuit has a second selection circuit, The second selection circuit has a function of switching the function of the input neuron circuit and the function of the hidden error circuit.
  11. 11. The semiconductor device according to claim 10, wherein, The error circuit has the function of generating a third signal from the second signal by converting a current into a voltage, When the error circuit is switched to the function of the output neuron circuit by the second selection circuit, the error circuit generates the error signal by multiplying the difference of the third signal and a supervisory signal by a differential coefficient generated based on the third signal, When the error circuit is switched to the function of the hidden error circuit by the second selection circuit, the error circuit generates the error signal by multiplying the difference of the reference voltage and the supervisory signal by a differential coefficient generated based on the third signal.
  12. 12. The semiconductor device according to any one of claims 1 to 4 and 6 to 9, wherein, The analog memory has a transistor that is connected to the memory, The transistor has an oxide semiconductor in a channel formation region.
  13. 13. The semiconductor device according to claim 12, wherein, The oxide semiconductor includes In, ga, and Zn.
  14. 14. An electronic device having the semiconductor device of any one of claims 1 to 4 and 6 to 9, wherein, The semiconductor device has a function as an encoder that encodes video data.
  15. 15. An electronic device having the semiconductor device of any one of claims 1 to 4 and 6 to 9, wherein, The semiconductor device has a function of obtaining a motion vector from one region to one region of a plurality of regions by comparing data of the one region extracted from first video data with data of the plurality of regions extracted from second video data.

Description

Semiconductor device and electronic apparatus The present application is a divisional application of patent application entitled "semiconductor device and electronic device" having an application number 201680003094.5 and an application date of 2016, 10 and 18. Technical Field One embodiment of the present invention relates to an electronic device or a semiconductor device including the semiconductor device. Note that one embodiment of the present invention is not limited to the above-described technical field. The technical field of the invention disclosed in the specification and the like relates to an object, a method or a manufacturing method. In addition, one embodiment of the present invention relates to a process, machine, product, or composition (composition of matter). Specifically, examples of the technical field of one embodiment of the present invention disclosed in the present specification include a semiconductor device, a display device, a light-emitting device, a power storage device, a memory device, a driving method thereof, and a manufacturing method thereof. In this specification and the like, a semiconductor device refers to an element, a circuit, a device, or the like that can operate by utilizing semiconductor characteristics. Examples of the semiconductor device include semiconductor elements such as transistors and diodes. Other examples of semiconductor devices are circuits including semiconductor elements. Another example of the semiconductor device is a device including a circuit including a semiconductor element. Background Models that mimic neural circuit networks and utilize computer-like processes, i.e., neural networks are expected to enable the fabrication of computers with higher performance than noerman computers. As shown in non-patent document 1, research and development in this field is active. In a neural network, units mimicking neurons are connected to each other by units mimicking synapses. The neural network can change the connection strength by learning, and can perform pattern recognition, associative memory, and the like for various input patterns at high speed. In pattern recognition, a method of judging consistency between the monitor data and the input data by using a unit imitating a neural network having a multi-layer sensor architecture as a pattern of a moving body to be studied for the monitor data and then supplying the image data as the input data, thereby extracting whether the moving body is included in the image is effective. The neural network with the multi-layered perceptron architecture is composed of neuronal circuits (cells mimicking neurons) and synaptic circuits (cells mimicking synapses). The synaptic electrical circuit needs to have a function of storing the connection strength between the neuron circuits, a function of multiplying the connection strength by the output of the neuron circuits, and a function of adding the multiplication results. Therefore, the abrupt circuit requires a memory for storing the connection strength, a multiplication circuit for realizing the multiplication function, and an addition circuit for realizing the addition function. With the increase in the size of television sets (TV), there is a demand for viewing high-definition images. Therefore, ultra High Definition TV (UHDTV) broadcasting is becoming practical. Japan, which has advanced UHDTV broadcasting, began a 4K broadcast service using a Communication Satellite (CS) and an optical fiber line in 2015. It is planned to start trial broadcasting of UHDTV (4K and 8K) by Broadcast Satellite (BS) in the future. Therefore, various electronic devices corresponding to 8K broadcasting are now being developed (see non-patent document 2). The practical broadcasting of 8K will be combined with 4K broadcasting and 2K broadcasting (full high definition broadcasting). In addition, image pickup devices are widely installed in various electronic devices such as digital cameras and mobile phones. As described above, UHDTV broadcasting has been put to practical use, and therefore, in recent years, the number of pixels of an image pickup device has also increased. As the number of pixels of the image pickup element increases, the amount of data obtained by image pickup inevitably increases. Therefore, data is required to be read and transferred at a higher speed. As a technique for coping with an increase in the amount of image data due to an increase in the number of pixels of an image pickup element, compression of image data is known. Patent document 1 discloses an image pickup device module that performs data compression by calculating difference data between previous image data and current image data at the time of capturing a video or continuous shooting. [ Reference ] [ Patent literature ] [ Patent document 1] Japanese patent application laid-open No. 2009-296353 [ Non-patent literature ] [ Non-patent literature ] 1] Yutaka Arima et al,"A Self-Learning Neural Network Ch