JPH04123019A - Spatial optical modulating element, driving method for the same, and neural network circuit - Google Patents

Abstract

PURPOSE:To obtain a spatial optical modulating element which can perform a threshold process at a high speed by connecting a part where plural photoconductive layers are connected in parallel electrically to a part where the transmission quantity of light is varied with an electric field. CONSTITUTION:A transparent conductive electrode 102 is formed on a transparent insulating substrate 101 and a photoconductive layer 103 which is separated by an insulating film 108 is laminated and a transparent conductive electrode 104 is further formed to sandwich a transparent conductive electrode 106 and a modulation part 107 with a transparent insulating substrate 105. When irradiated with incident light 109, the photoconductive layer 103 decreases in resistance RP1, a voltage V is applied mainly to the CL and RL of the modulation part 107, and read light 110 is transmitted through the modulation part 107 to obtain output light 111. At this time, the intensity of the incident light needs to the larger than a certain value so as to obtain the output light, so the element operates as a threshold element and this threshold value can be varied with the voltage V. Consequently, the spatial optical modulating element which can performs the threshold value processing by a fast operation is obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION Industrial fields of application The present invention relates to a spatial light modulation element used in an optical processing device or a projection type display, a method for driving the spatial light modulation element, and an input/output animal similar to the nervous system, e.g. Pattern recognition, associative memory, and related memory are related to neural network circuits that perform parallel calculation processing. There are reasons such as the parallel nature of light is compatible with neural networks that perform calculations using parallel dynamics, and optical connections that do not require wiring like electrical circuits make it easy to realize multiple connections between neurons. Although various configurations of optical neurocomputing have been proposed, one of the basic calculations of neurocomputing is to perform summation using a large number of human power and perform threshold processing on the result. Kaamo: It is most important to realize a spatial light modulation element with such arithmetic functions in order to implement a neural network into hardware. There have been few proposals for optical neurocomputers using this, and patent application No. 63-298701.
Patent application Hei 1-78820. Patent application Hei 1-18639
3, these ζ飄 are photo-threshold elements in which a plurality of photoconductive layers are arranged in series and electrically connected to a liquid crystal layer, or a plurality of photoconductive layers arranged in series and a field effect. 10(a) and (b) show the problems to be solved by the present invention, which relate to an optical threshold device in which a type transistor and a liquid crystal layer are electrically connected, and a neural network constructed using these optical threshold devices. shows an equivalent circuit of the optical threshold device mentioned in the conventional example. FIG. 10(a) shows the former of the conventional example, (b)
) is the latter. The examples in Figures 10(a) and (b) both show that the amount of light incident on each photoconductive layer connected in series corresponds to the photoconductive layer connected in series according to the law of partial pressure. Even if optical summation is realized using the function of changing the voltage applied to the entire layer, the electrical resistance of the liquid crystal layer is large,
Furthermore, the dark resistance of the photoconductive layer is as thick as that of the liquid crystal layer. The combined resistance becomes very large. Therefore, the current flowing through the circuit is very small, and it takes time of the order of Ct seconds or more to reach a steady state in which the voltage is distributed according to the law of voltage division. In addition, it has been difficult to perform high-speed calculations using conventional optical threshold elements and neural networks using them.Spatial light modulation with optical summation and threshold processing functions Although threshold processing is performed using electronic circuits or computers, it is reported that it is caused by the hierarchical structure of neural circuits. If there is no spatial light modulation element that performs optical summation and optical threshold processing, in order to realize this hierarchical network, after performing optical-to-electrical conversion for thresholding, further electrical-to-optical conversion is required. (t) There was a problem that the parallelism of light could not be taken advantage of, effectively reducing the calculation speed. Spatial light modulators have memory properties and are attracting attention as useful devices that can realize functions such as image inversion and coherent/incoherent conversion. The first and second inventions of the present application are intended to solve the above-mentioned problems of the conventional sumo wrestlers who sometimes introduce a PN junction into the conductive layer. The third invention of the present application provides a spatial light modulator that can perform optical summation and threshold processing at high speed, and the third invention prevents deterioration of the spatial light modulator that is made of a photoconductive layer having rectifying properties and a ferroelectric liquid crystal. This invention provides a method for driving a spatial light modulation element to extend its lifespan, and the fourth to sixth inventions of the present application (Providing a neural network circuit that can learn and easily realize a hierarchical network model) Means for Solving the Problem The spatial light modulator of the first invention (a part in which a plurality of photoconductive layers sandwiched between highly conductive electrodes are connected in parallel and the amount of light transmitted by an electric field) 4 Spatial light modulation element ζ of the second invention
y, comprising a portion in which the amount of light transmitted changes depending on an electric field, a photoconductive portion, and a field effect transistor, the photoconductive portion being electrically connected to the field effect transistor, and the photoconductive portion being electrically connected to the field effect transistor; The photoconductive portion is characterized in that a plurality of photoconductive layers sandwiched between conductive electrodes are connected in parallel.

A method for driving a spatial light modulator according to a third aspect of the present invention (hereinafter referred to as
The photoconductive layer has positive and negative pulse components of less than T/2 between T) and has a direct current component that provides a reverse bias to the photoconductive layer across T. Neural network circuit ζ of invention No. 4 A part in which multiple photoconductive layers sandwiched between conductive electrodes are connected in parallel is electrically connected to a part in which the amount of light transmitted changes depending on the electric field, and these parts are made of transparent insulating material. The neural network circuit of the fifth invention is characterized by comprising a spatial light modulation element, an optical mask, and a light emitting element sandwiched between substrates.
A field-effect transistor includes a photoconductive portion formed on a transparent insulating substrate, a portion whose amount of light transmission changes depending on an electric field, and a field-effect transistor in which the photoconductive portion is electrically connected to the field-effect transistor. A spatial light modulator, an optical mask, and a light emitting element, each of which is connected in parallel with a plurality of photoconductive layers sandwiched between partial force conductive electrodes having photoconductivity. The neural network circuit of the sixth invention includes a transparent conductive electrode, a photoconductive layer, and an array of conductive electrodes formed on a transparent insulating substrate, and further includes a portion where the amount of light transmitted changes depending on an electric field, and a transparent conductive layer. It is characterized by comprising a spatial light modulator having a structure in which electrodes are stacked, an optical mask, and a light emitting element.

A voltage is applied from the outside in a structure in which a part in which multiple active photoconductive layers are connected in parallel and a part in which the amount of light transmitted changes depending on the electric field (hereinafter abbreviated as the modulation part) are connected in series. Case: When the photoconductive layer is not irradiated with light! The electrical impedance of the photoconductive layer is large, and the external voltage is mainly applied to the photoconductive layer.When light with an intensity above a certain level is incident on the photoconductive layer, the electrical impedance of the photoconductive layer decreases. An external voltage is now applied to the modulation part, which switches the light that passes through the modulation parta.In other words, the optical threshold processing function can be realized.In addition, the photocurrent flowing through the photoconductive layer increases almost in proportion to the amount of light4.
Because the individual photoconductive layers are connected in parallel, the current flowing through the entire photoconductive layer is proportional to the total amount of light incident on each photoconductive layer, and the charge that is charged to the modulated area of the skin over a certain period of time is The quantity is proportional to the current. Therefore, it is possible to realize optical summation. However, when the operating voltage or current of the modulation section is too large to connect to the photoconductive layer, the photoconductive layer is connected to the gate electrode of a field effect transistor (hereinafter abbreviated as FET). Effects similar to those described above can be achieved by connecting the modulation portion to the electrodes.

When the threshold processing function of neurons is realized using such a spatial light modulation element, the intensity of light incident on the photoconductive layer, in other words, the threshold value, at which the voltage begins to be applied to the modulation area, can be controlled by an external voltage applied to the spatial modulation element. By controlling the external voltage, learning of the neural network circuit can be converged efficiently.Also, by arranging multiple spatial light modulation elements in parallel, threshold processing can be performed in parallel. These spatial light modulators can also operate in transmissive mode, alternating with optical masks corresponding to synaptic connection patterns. Although it is possible to easily create a hierarchical neural network by simply overlapping each other, there is a method that alternately applies positive and negative voltage pulses to drive a spatial light modulator composed of a ferroelectric liquid crystal and a photoconductive layer with rectifying properties. Due to the rectifying properties of the photoconductive layer used, the current is equivalent to a state in which a direct current bias is always applied to the liquid crystal layer. Therefore, the liquid crystal becomes more likely to decompose, resulting in a shortened lifespan.

Therefore, by applying a DC offset voltage superimposed on the positive and negative voltage pulses in order to cancel out this DC bias, it is possible to suppress the decrease in life.Also, when such an element is used as an optical threshold element, this offset voltage ,
It is possible to change the slope of the threshold value characteristic and also to control the threshold value.A very effective embodiment of the present invention will be described with reference to the drawings. 1 shows an embodiment of a spatial modulation element according to the invention.

FIG. 1 shows an example of a transmission type, in which (a) is a cross-sectional view, and (b) is an equivalent circuit diagram. (Example ζ; U ITO or S n O,) 10
2, and a photoconductive layer 103 is laminated thereon.
Furthermore, a transparent conductive electrode 104 is formed, and a transparent conductive electrode 106 and a modulation portion 1 are formed using this and a transparent insulating substrate 105.
Even if 07 is sandwiched, however, the photoconductive layer 103ζ
The operation of the spatial light modulator even when separated by the insulating film 108 will be described using the cross-sectional view of FIG. 1(a) and the circuit diagram of FIG. 1(b).
Even though the AC voltage V is applied to the RPI (Li=1...n, n: total number of photoconductive layers) and the CL and RL of the modulation section 107, in this state the voltage Vial
CL, RtJ: remote cp+, Rp+i: mainly cam Here, when the photoconductive layer 03 is irradiated with incident light 109, the resistance RPI of the photoconductive layer 103 decreases, and the voltage V is mainly applied to CL and RL. Therefore, the readout light 110 passes through the modulation portion 107 and the output light 111
Therefore, the output light intensity depends on the incident light intensity, and although it functions as an optical writing type and transmission type spatial light modulator, the incident light intensity must be above a certain value ζ to obtain the output light. Therefore, it also functions as a threshold element.
In addition, this threshold value can also be changed by the voltage V when using the spatial light modulator shown in FIG. 1 as a reflective type.
A dielectric mirror or A with multiple layers of dielectric material stacked between
A metal thin film mirror with high reflectance such as g, AI, Cr, Ni, or Mo is formed on the transparent conductive electrode pattern 104 instead of Ag, Al, Cr, or N.
A metal thin film with high reflectivity such as i, Mo may be used instead.
y -X In order to increase the resolution, a thin film made of a material having a band gap sufficiently smaller than that of the photoconductive layer 103 is placed between the photoconductive layer 103 and the transparent conductive electrode pattern 104 as a light absorption layer that prevents reflection of the incident light 109. The operation of the reflective spatial light modulator is basically the same as that of the transmissive type shown in Figure 1, but the readout light enters from the transparent insulating substrate 105 side and is output. Light is transparent insulating substrate 1
The first invention, the spatial light modulator, emits light from the 05 side.
Similar operation can be obtained in both transmission type and reflection type cases. Here, the modulation part 107L KDePO4,B 14
If a liquid crystal is used for the modulation section 107, which is made of a material such as a ferroelectric crystal such as Ti2Q+t or a material whose light transmission amount changes depending on an electric field, such as a liquid crystal, a nematic liquid crystal such as ζ may be used as the liquid crystal material. A chiral smectic C phase ferroelectric liquid crystal may also be used (~ In this case,
It is more effective because the thickness of the liquid crystal layer can be reduced, the capacitance can be increased, high-speed response is possible, and it also has a memory function.

The material used for the photoconductive layer 103 acts as a dielectric in the dark, but loses its dielectric properties due to photoconductivity when irradiated with light. For example, ficds, CdTe, and Cd.
Se, ZnS, Zn5e, GaAs.

Compound semiconductors such as GaN, GaP, GaAlAs, InP, etc. Amorphous semiconductors such as Se, 5eTe, AsSe, SLG e, S I +-xcx,
S l +−xG ex.

Polycrystalline or amorphous semiconductor of G e 1-++ Cx (0<x<1) or (1) Phthalocyanine pigment (abbreviated as Pc) such as metal-free Pc, XPc (X=Cu,
Ni, Co, Tie, Mg, Si
(OH)2, etc.), AlClPcCl, Ti0CI
PcC1, InClPcC1, InClPc,
InBrPcBr etc', (2) Monoazo dye
Azo-based pigments such as disazo dyes (3) Penylene-based pigments such as penylene acid anhydride and penylene acid imide (
4) Indigoid dyed porridge (5) Quinacridones It,
(6) Anthraquinones Polycyclic beef nons such as pyrenequinones (7) Cyanine dark in color (8) Beef sustenten dyeing
(9) Charge transfer complexes such as PVK/TNF; (10) Eutectic complexes formed from beryllium salt dyes and polycarbonate resins; (11) Organic semiconductors such as azulenium salt compounds; and amorphous Si, Ge, S il-xC
x.

SII-XGeX, Ge1-xCx
(Hereinafter, a-8i, a-Ge, a-8il-*cX
, a-8il-xGex, aGe+-++C,
When using hydrogen or a halogen element in the photoconductive layer 103 (abbreviated as B and AI, which are p-type impurities, are used for control.
, Ga, or n-type impurities such as P and As.
, Sb, and other elements may be added (°C) By stacking amorphous materials doped with impurities in this way, p/n, p/i
, i/np/i/n junctions, etc. may be formed to form a depletion layer within the photoconductive layer 103 to control the dielectric constant, dark resistance, or operating voltage polarity (by doing so, This state can be quickly released by not only using such an amorphous material but also by stacking two or more of the above materials to form a Peter junction and depleting the photoconductive layer 103. You can also form layers (t
The thickness of the photoconductive layer 103 is preferably 0.01 to 100 μm, and since light mainly passes through areas where there is no photoconductive layer 103, the intensity of transmitted light is not reduced as much as possible.The area of the photoconductive layer 103 is a transparent conductive layer. It is preferable that the area of the electrode pattern 104 is 75% or less.
8 It S IOx, S iNX, A l
Materials such as 203° polyimide are effective. FIG. 2 shows a circuit diagram of an example of a spatial light modulator according to the second invention. This spatial modulation element basically consists of two FETs (FE
TI, FET2), a modulation part (represented by resistance component RL and capacitance by CL), and a plurality of parallel-connected photoconductive layers (also represented by R1 and C1, where 1, i=1...n
, n: total number of photoconductive layers).
The operation of the spatial light modulator is explained using this circuit diagram even though it is connected between the gate electrode of ET2 and the ground.4 When v6 is applied to the gate electrode of FETI, FL...
FETI conducts and the photoconductive layer CPI is charged (however,
It is assumed that R1 in the dark is very large4). When charging of Cp+ is completed, Vo is reduced to cut off conduction of FETI. At this time, the gate voltage of FET2 is approximately equal to VD, and FET2 is conductive and the alternating current voltage V is applied to the modulation portions RL and CL.
is applied. Next, the photoconductive layer CPI is irradiated with light, and when the light intensity becomes large enough to sufficiently reduce the resistance Rpl of the photoconductive layer, the charges stored in the photoconductive layer are neutralized.
The gate voltage of FET2 decreases. Sota & F
The conduction of ET2 is cut off, and the alternating current voltage V is no longer applied to the modulation part CL.In this way, by irradiating the photoconductive layer with light, the voltage applied to the modulation part can be controlled, creating a spatial light modulation element with a threshold processing function. It can be operated as ama7',,,FET2
The speed at which the gate voltage decreases is proportional to the current flowing through the entire photoconductive layer, so even though optical summation can be achieved, the FET is usually constructed from a thin film TPT. Not only i,, a-3i:H but also polycrystalline silicon, polycrystalline GaAs, aSll are used as the semiconductor layer constituting the TFT.
-*Cw :H, a-3i+-xGex :H(0<
x<1), etc. are used.

Further, the photoconductive layer and the modulation portion are made of the above-mentioned materials. FIG. 3(a) shows an example of a circuit for explaining the method for driving the spatial light modulation element of the third invention. Figure 3(b),
(c) shows the waveform of the power supply voltage V.

A photoconductive layer with rectifying properties is represented by a parallel circuit of a diode and a capacitance C-, and a ferroelectric liquid crystal is represented by a capacitance C.

Tsuie: As a waveform for driving the ferroelectric liquid crystal, a voltage waveform as shown in FIG. 3(C) is used, which takes advantage of the memory properties of the ferroelectric liquid crystal. This waveform is an AC waveform consisting of positive and negative pulses with a pulse width of less than T/2 during one period (time T), and has no DC component.
If the waveform shown in Fig. 3(c) is used as the power supply voltage shown in Fig. 3(a), a positive potential will always be generated at point A between C# and CL due to the rectifying property of the photoconductive layer. That is, a direct current bias is always applied between the terminals of CL, that is, to the ferroelectric liquid crystal cell, which accelerates the deterioration of the liquid crystal.

Therefore, deterioration of the liquid crystal can be prevented by using the waveform of FIG. 3(b), which is obtained by superimposing a negative DC component that cancels this positive potential on the waveform of FIG. 3(c), as the power supply voltage. However, if the direction of the diode in Figure 3 (a) is opposite, it would be better to superimpose a positive DC component on the waveform in Figure 3 (c) (that is, with respect to the plate diode). All you have to do is to superimpose the DC component so that it becomes a reverse bias.Also, the number of positive and negative pulses can be the same or different, and the heights of the positive and negative pulses can also be the same and L different. A specific example will be described below.Example 1 As shown in the cross-sectional view of FIG.
．． ITO with a thickness of 1 to 0.5 μm is formed by sputtering to form a transparent conductive electrode 402. Next, plasma C is applied.
A photoconductive layer 403 was fabricated by laminating a p/i/n diode structure a-8i:H film with a thickness of 1 to 3 μm using the VD method and forming a pattern.
After forming a 5iN- pattern with the same film thickness as 5i:H, a TO electrode pattern 40 with a thickness of 0.1 to 0.5 μm was formed.
Orientation III where 5 was formed and 9 was then subjected to rubbing treatment.
! A ferroelectric liquid crystal layer 410 with a thickness of 0.5 to 3 μm is placed between this and a glass substrate 409 on which an ITO transparent conductive electrode pattern 407 and an alignment film 408 are laminated.
A polarizer 411 and an analyzer 412 are placed on both sides of this cell to fabricate a spatial light modulator 413. The operation was confirmed by applying an alternating current voltage and using white light as the incident light 414. As a result, the change in the output light 415 with respect to the incident light 414 showed a threshold characteristic, and light summation was performed. We have confirmed that this light summation and light threshold processing can be performed 10~10°000 times per second. , the threshold value should be set by increasing the pulse height of the AC voltage.
In addition, this spatial light modulator also functions as an optical logic element.For example, for simplicity, the number of photoconductive layers 403 is reduced to 2.
If light is not incident on two photoconductive layers 403 at the same time, no voltage will be applied to the liquid crystal layer 410, resulting in an AND circuit, and if light is incident on even one photoconductive layer 403, a voltage will be applied to the liquid crystal layer 410. This results in an OR circuit. Furthermore, by adjusting the polarizer 411 and analyzer 412, it also functions as a NAND or NOR circuit.

Example 2 A reflective spatial light modulator 501 as shown in FIG. 5 was manufactured. This jL was obtained by replacing the ITO electrode pattern 405 of the transmissive type shown in FIG. 4 with an AI electrode pattern. Furthermore, a polarizer 502 and Analyzer 503 is the fifth
Arranged as shown in the figure, input the readout light 505 from the opposite side to the incident light 504, and observe the output light 506 from the beam splitter = 506. As a result, the light sum and light threshold characteristics were observed in the same way as in Example 1. Embodiment 3 FIG. 6 shows a schematic cross-sectional view of a spatial light modulation element 601 according to an embodiment of the present invention. This spatial light modulator is a-3i:H
Two TPTs using the semiconductor layer 602 are formed on one glass substrate 603. To form TPT, first, a transparent ground electrode 604 and a gate electrode 605 made of ITO or NESA are formed on a glass substrate 603 using, for example, Cr.
After that, a gate insulating film 60 and a semiconductor layer 602 are formed.
- A semiconductor protective layer 607 was formed and patterned by plasma CVD, and an n-type semiconductor layer 608 was interposed to improve ohmic properties, and then a source electrode 609 was formed.
- Batch formation of the drain electrode 610 using AI, etc.
Next after fabricating TFT! A method of forming a hole in the gate insulating film 606 by etching, forming and patterning a photoconductive layer 611 of, for example, a-3i.H, and then forming an electrode 61.
Then, a pixel electrode 614 for applying an electric field to the ferroelectric liquid crystal 613 is formed with a transparent electrode such as ITO. Orientation M61
A counter electrode 617 and a light-shielding layer 618 are installed on the other glass substrate 616.A ferroelectric liquid crystal 613 is then applied between both plates. The change in transmitted light intensity with respect to the incident light intensity irradiated from the glass substrate 603 side with TPT formed on the photoconductive layer 611 of this spatial light modulation element in which color polarizing plates 620 are placed before and after the substrate shows a threshold characteristic. , and after confirming that light summation can be performed for the incident light, the pixel electrode 614 was replaced with an electrode made of Al, Cr, etc.
By arranging this spatial light modulator in an optical system as shown in FIG. 5, it can also operate as a reflective spatial light modulator. A transparent electrode 702. P/I/N structure a-8i:H photoconductive layer 70th order ferroelectric liquid crystal layer 70 East A spatial light modulation element was fabricated in which a transparent electrode 705 and a glass substrate 706 were sequentially laminated.This process is shown in Figure 3(c). The memory property of the ferroelectric liquid crystal 704 was lost after approximately 10 degrees when a voltage waveform like the one shown in Figure 3(b) was applied. Furthermore, we confirmed that the sensitivity of the spatial light modulator to the incident light can be changed by controlling the voltage of the DC component shown in Figure 3(b). As layer 703, a-
It is possible to obtain the same effect as L by using the above-mentioned material instead of 3i:H to form a photoconductive layer having rectifying properties, and that the p/i/n structure a-5i:H photoconductive layer 703 and strong Example 5 To confirm that the same effect can be obtained when a dielectric mirror is inserted between the dielectric liquid crystal layers 704, a-3i
A conductive electrode array 707 made of A1 or Cr, for example, is formed between the :H photoconductive layer 703 and the ferroelectric liquid crystal layer 704 to fabricate a reflective spatial light modulator. Shown in b).

When a voltage waveform as shown in FIG. 3(b) is applied to this spatial light modulator, the voltage waveform shown in FIG. 3(c) is 2,
It was confirmed that the life of the element was extended by more than 1,000 times.
7h (a) schematically shows the configuration of the neural network circuit, (b) shows the -ff1J pattern stored in the circuit, (c) shows multiple input patterns & (d) is an example of an incomplete pattern used in the experiment. A surface light emitting device 801 consisting of a two-dimensional array of light emitting diodes, electroluminescent (EL) elements, or fluorescent lamps and a surface diffuser plate, etc.
and two active matrix liquid crystal cells (hereinafter abbreviated as AM-LC cells) using an a-3i:H or polycrystalline Si transistor array as an active element and a 90° twisted nematic liquid crystal as the liquid crystal. M
-LC(1)802. A M -L C (2) 80
3) Just L AM-LC arranged in parallel
(2) 803 is an optical mask element 80 corresponding to the synapse
2, and it is designed to be learnable. Ma? The number of pixels of QAM-LC cell 802.803 is 400x
400 (-160000) pieces, each operated by a drive circuit 804.805.Furthermore, in the spatial light modulation element fabricated in Example 2, the number of photoconductive layers was increased to 16, and the number of photoconductive layers was 4g vertically and 4 horizontally. AML
C cell (1) 802. (2) Placed parallel to 803
The principle of operation of the neural network circuit (here we take up associative memory, which is one of the functions of the Cyo nervous system) will be explained using FIG. 8.

4x4 (=16) matrix-like input cover pattern (X
l), for example, one of the patterns shown in FIG. 8(b) (for example, the top pattern) is imaged with a television camera 807.
It is processed by the computer 808 and the eighth
Displays a 16x16 (=256) multiple image (XIL) pattern as shown in figure (c) L AM-L C (2)
Display 18x16 (=256) memory patterns (M) on 803 and arrange them in a one-to-one correspondence with the elements ζ of the matrix of M.
An image is formed so that one element of L and M is focused on one cell including one photoconductive layer of the spatial light modulator 706.The readout light of the spatial light modulator array 806 is an Ar laser or a He- Using a Ne laser (threshold-processed reflected image was captured with a TV camera at 8°9), the recall result Y+(4x4
If this YI and the input The initial state is a 0 matrix, and for all other patterns, the output Y is the input X
Learning is repeated until the memory pattern M and I was able to recall Xl.
In addition, we were able to recall complete patterns from incomplete input patterns for other patterns.In this way, we formed a circuit using a spatial light modulation element with a threshold processing function, and used the associative memory of the nervous system. Kokote, AM-L, C cell (1) 802. (2)803
4. In addition, a voltage waveform as shown in FIG. 3(b) was applied to the spatial light modulation element array 806. We confirmed that the convergence of learning was improved by controlling the threshold by changing the negative pulse voltage height and DC component voltage.We also found that the recognition rate for incomplete pattern input was improved. Furthermore, the matrix size of the human covertane is increased, and the number of photoconductive layers included in the spatial light modulation element is increased accordingly, and the neural network circuit is configured in the same manner as above.More and more complex patterns It was also confirmed that associative memory can be performed even for 4x4 images using the reflective spatial light modulator of Example 3.
It was confirmed that the same effect as above was obtained by constructing a matrix spatial light modulation element array 8°6.Also, a 4x4 matrix spatial light modulation element was constructed using the reflective spatial light modulation element fabricated in Example 5. array 806
It has been confirmed that the same effect as above can be obtained by configuring L. By forming a lens, it is possible to reduce the decrease in light intensity due to light scattering and optical loss talk, and if learning converges faster, good output results can be obtained even for incomplete patterns. PLZ uses a liquid crystal, which is one of the devices that changes the amount of light transmitted by an electric field, as an optical mask corresponding to the synapse.
T, KD2POJ, BiaTi 20+1, etc. may be used, or a photochromic element that changes the amount of light transmitted by light irradiation may be used (- Also, in the case of a neural network circuit that does not perform learning, optical Although the mask is not rewritable, a fixed mask fixed with silver salt or dye or made by vapor deposition of metal is used.In this case, the contents of the fixed mask are synapses obtained by computer simulation of learning. Even if this value is included, if this fixed mask is outside the spatial light modulator array 806
If the element size is smaller than the thickness of the glass of the spatial light modulation element array 806, the light scattering will cause a decrease in the light intensity and optical loss talk. In order to prevent such a situation, it is effective to form a fixed mask within the spatial light modulation element array 806.

Specifically, there is a gap between the glass substrate on the incident light side and the transparent conductive electrode, or between the transparent conductive electrode on the incident light side and the photoconductive layer.Spatial light modulation It is more effective to add the above-mentioned lens to the glass substrate of the element array 806. Also, it is confirmed that by such a method, it is possible to obtain exactly the same results as the simulation results. Example 7 Example 1 First, a neural network circuit 901 based on a hierarchical neural network model as shown in FIG. 9(a) is configured using the transmissive spatial light modulator shown in FIG. , 44x4 matrix input information (X) that explains the configuration of this circuit.
LA is displayed on the surface light source 902.A multiplexed image (XL) is imaged on the AM-LC (1) 904 using the lens array 903 arranged in a 4x4 matrix, as in the fifth embodiment. Corresponds to the nerve cells in the layer (in this case, the number of nerve cells is 16) L
AM-LC (1) 904 is used as a rewritable optical mask corresponding to the synapse between the input layer and the intermediate layer. (1) The two polarizers sandwiching 904 are color polarizers whose polarization directions are parallel, and polarize light at a specific wavelength (for example, λ1).
-Xλ2 wavelength is almost transmitted. Therefore,
The light of λ1 is modulated by the A M -L C(1) 904, but the light of λ2 is transmitted regardless of the operating state of the liquid crystal layer. A spatial light modulator array in which the transmissive spatial light modulators of Example 1 (the number of photoconductive layers connected in parallel per element is 4x4=16) is arranged in a matrix of (
(hereinafter abbreviated as SLM array) (1) 905 is irradiated with light. Corresponds to neurons (in this case, the number of neurons is 16).SLM array (1) 9
When passing through the color filter 906, the λ1 configuration is blocked and only the λ2 configuration is transmitted. A M-L in a 16x16 matrix display used as a rewritable optical mask corresponding to the synapses of
M A M -L C(2)9 incident on C(2)908
The transmitted light of 0.08 is incident on the reflective 4x4 matrix SLM array (2) 909 (the number of photoconductive layers per element is 4x4 = 16) shown in Example 2. Here,
The photoconductive layer of the SLM array (2) 909 is made of a material that absorbs light with wavelength λ2 well, and the SLM array (2) 909 corresponds to the neurons of the output layer (in this case, the number of neurons is is 16 pieces).SLM array (2) 9
The output light from 09 passes through a beam splitter 910 and is received by a photodiode array (hereinafter abbreviated as PDA) 911 arranged in a 4x4 matrix.
SLM array (1) 905 and SLM array (2) 9
09fi It performs optical summation and outputs the result of threshold processing on the result.4 P D A
Using the electronic computer 912, the signal from 911 calculates the amount of synaptic correction using a feedback type supervised learning method (for example, the error backpropagation learning method), and
904. (2) This system has undergone 908 learning (input layer - 16 neurons, intermediate layer - 16i, output layer - 16 neurons).

This neural network circuit 901 uses outputs from optical sensors that respond to 16 types of single wavelength light as input information (for example, input information about sunlight, fluorescent lamps, incandescent lamps, sodium lamps, and candle flames). Acquisition SLM was trained to recognize and distinguish between these.However, during learning SLM
Array (1) 905. (2) As a result of changing the threshold value by changing the voltage applied to 910, a recognition rate of 100% was obtained for the input information used for learning, and for other inputs that were not used for learning. A recognition rate of 99.5% or more was obtained.In this case, the convergence degree of learning was 5 to 1 L compared to the case where no threshold learning was performed.
In this way, the hierarchical neural network is now 0 times faster.
＼If the materials of the photoconductive layer used in the SLM array are made up of materials with smaller inhibitor widths in the direction in which light travels, the AM
- It can be realized very easily by simply stacking transmissive spatial light modulators such as LC and SLM arrays alternately. Also, AM-L in this neural network circuit 901
A similar effect can be obtained by increasing the matrix size of the C and SLM arrays beyond this example.
- In this case, the functionality can be further improved by increasing the number of pairs of LC and SLM arrays and increasing the number of layers.Similar results can be obtained by using the spatial light modulator arrays of Examples 3 and 5 as in Example 7. It was confirmed that similar results were obtained by using other optical masks used in Example 7 instead of -AM-LC.SLM array (1) 905. (2) When a lens is formed on the 909 glass substrate, learning converges more efficiently and the recognition rate improves.Also, the learning results obtained in the simulation were made into a fixed mask, and A M -L C (1) 904. (2
) 908 instead of c<SLM array(1) 905.
(2) Effects of the Invention The present invention provides high-speed operation without obtaining simulation-like results when inserted between the glass substrate of 909 and the transparent conductive electrode or between the transparent conductive electrode and the photoconductive layer. This not only makes it possible to obtain a spatial light modulator capable of optical summation and threshold processing, but it also provides a driving method that makes it possible to extend the lifetime of the spatial light modulator.It also provides excellent learning convergence and facilitates the creation of hierarchical network models. Although the neural network circuits that form can be obtained

[Brief explanation of the drawing]

FIGS. 1(a) and (b) IL are a cross-sectional view and a circuit diagram of an embodiment of a spatial light modulator according to the present invention, respectively. FIG. 2 is a circuit diagram of a reflective spatial light modulator according to an embodiment of the present invention. Figure 3 (a), (b) and (C)
are respectively circuit diagrams showing circuits used in an embodiment of the method for driving a spatial light modulation element of the present invention.A voltage waveform detector of a conventional example.A voltage waveform detector of the present invention.Figure 4 is an embodiment of the spatial light modulation element of the present invention. Cross-sectional view of an example Figure 5C Cross-sectional view of an embodiment of the reflective spatial light modulator of the present invention Figure 6 Upper Cross-sectional view of an embodiment of the spatial light modulator of the present invention Figure 7(a) , (b) i Cross-sectional diagram of the spatial light modulation element in the embodiment of the present invention (a), (b), (c)
and (d) (respectively) Schematically illustrating the configuration of an embodiment of the neural network circuit of the present invention; Diagram illustrating an example of a pattern stored in the circuit; Diagram illustrating an example of a multiplex image of an input pattern; and A doctor who shows an example
FIGS. 9(a) and (b) 4;L respectively. FIG. 9(a), (b) 4; a), (b
) i;& is a circuit diagram of a conventional spatial light modulation element. 101, 105...Transparent insulating substrate, 102.104
．． 106...Transparent insulating electrode 103...Light guiding member 107...Modulation part, 108...Insulating wind 109...
...Incoming light 110...Reading light 111...Output light 401.409...Glass base K 402,4
07...Transparent conductive outer capacitor K 403...Light guide heavy-on
404...Insulation [205...IT O11i polar pattern, 406.408...Orientation K 410.
... Ferroelectric liquid a411... Polarizer, 412... Analyzer, 413... Spatial light modulation element, 414... Incident light 415... Output light 601... Spatial light modulation element, 6
02...Semiconductor 603,616...Glass substrate,
604... Earth electrode 605... Gate electrode 60
6...Gate insulation wind 607...Semiconductor protection 60g
. . . N-type semiconductor terminal 609 . . . Source electrode 610 .
・Drain electrode 611... Light guide layer 612... Electrode 613... Ferroelectric liquid crystal 614... Pixel electric machine
615,619...Oriented wind 617...Counter electrode 6
18...Shading effect 620...Color polarizing plate, 701.
706...Glass substrate, 702.705...Transparent electrode 703...Photoconductive W#704...Ferroelectric liquid crystal 7
07... Conductive electrode array, 801... Surface emitting device, 802... AM-LC (1), 803... AM-
LC (2), 804.805... Drive same section 806...
・Spatial light modulation element array, 807.809...TV camera, 808...Calculation a 901...Neural network apposition 902...Surface light #903,907...
Lens array, 904...AM-LC (1), 905
...SLM array (1), 906...Color filter 908...AM-L C (2), 909...S
LM array (2), 910...beam splitter-9
11...Photodiode array, 912...Electronic computing Name of agent: Patent attorney Akira Okaji and 2 others otto
s Encounter θ1 Absolute slowness life law area (N shi diagram (Ql 4o! 40q 402.407 Upward 3 Peeking 408 1O No Las Base Section Transparent Li'Tl Loss Electricity - Light 1lItL4 Sewing t4Ill I TQtIf Turn Orientation 5!! Hu Dengou Ching 1 & Tako 4) Δ゛-2 (11th two. 4I5-- Out 7] Light 4/1 Sky Ran Light Transformation Photon Dungeon Incident Light t1! Kaden Out 73 Light theory light 1 ↓ 1 ＼−−−～−−−ノFigure (C) rd) No. C

Claims (11)

[Claims] (1) A spatial light modulation element characterized in that a portion in which a plurality of photoconductive layers sandwiched between conductive electrodes are connected in parallel and a portion in which the amount of light transmitted changes depending on an electric field are electrically connected. (2) The device includes a portion where the amount of light transmitted changes depending on an electric field, a photoconductive portion, and a field effect transistor, and the photoconductive portion is electrically connected to the field effect transistor. , and the photoconductive portion is constructed by connecting a plurality of photoconductive layers sandwiched between conductive electrodes in parallel. (3) The spatial light modulator according to claim 1 or 2, wherein the photoconductive layer has rectifying properties. (4) The spatial light modulator according to claim 1 or 2, wherein the portion whose amount of light transmission changes depending on the electric field is made of ferroelectric liquid crystal. (5) In a spatial light modulator having at least a photoconductive layer having rectifying properties and a ferroelectric liquid crystal, the applied voltage waveform has positive and negative pulses of less than T/2 during one period (hereinafter referred to as T). 1. A method for driving a spatial light modulator, comprising: a direct current component having a direct current component that is reverse biased to the photoconductive layer over T. (6) Spatial light modulation in which the part where multiple photoconductive layers are connected in parallel between conductive electrodes and the part where the amount of light transmitted changes depending on the electric field is electrically connected, and these are sandwiched between transparent insulating substrates. A neural network circuit comprising a device, an optical mask, and a light emitting device. (7) A part having photoconductivity formed on a transparent insulating substrate, a part whose amount of light transmission changes depending on an electric field, and a field effect transistor, wherein the part having photoconductivity is provided in the field effect transistor. are electrically connected to each other, and the photoconductive portion comprises a spatial light modulator, an optical mask, and a light emitting element, each of which has a plurality of photoconductive layers sandwiched between conductive electrodes connected in parallel. neural network circuit. (8) It has a structure in which a transparent conductive electrode, a photoconductive layer, and an array of conductive electrodes are formed on a transparent insulating substrate, and a part where the amount of light transmitted changes depending on an electric field and a transparent conductive electrode are laminated. A neural network circuit comprising a spatial light modulator, an optical mask, and a light emitting element. (9) The neural network circuit according to any one of claims 6 to 8, wherein the optical mask is composed of a spatial light modulation element whose amount of light transmission changes depending on an electric field or light irradiation. (10) Any one of claims 6 to 8, wherein the optical mask is formed on the insulating substrate of the spatial light modulator, in the insulating substrate, or between the transparent conductive electrode and the photoconductive layer. Neural network circuit described in. (11) The neural network circuit according to any one of claims 6 to 8, wherein a lens is formed on the transparent conductive substrate of the spatial light modulation element.