JPH0498228A - Optical computing element - Google Patents

Abstract

PURPOSE:To obtain an laminated optical computing element provided with a high density, large capacity and rapid spatial light modulating element by processing respective components values of a vector outputted from a photodiode array by respective thresholds. CONSTITUTION:All the column components of light outputted from an optical valve 2 are converged into corresponding light receiving elements in a photodiode array 3 and matrix-vector products can be obtained from the output of the array 3. Since operation can be completely optically executed, extremely rapid processing can be executed in parallel. An output signal photoelectrically converted by the array 3 is threshold-processed by a comparator 4 and fed back to an LED array 1. Even when an input vector includes incomplete information, the most similar information is selected from plural complete information stored in the matrix of the optical valve 2 and an complete output vi<(infinity)> is obtained. When the optical valve 2 using a monocrystal thin film layer is utilized, a large capacity matrix can be attained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an integrated chip-type optical arithmetic element having a laminated structure, and particularly to a multiplier that optically performs vector and matrix multiplication. This type of optical arithmetic unit is used as a component of a neural network that imitates the neural network of living organisms, and is capable of performing associative functions, image recognition functions, etc.

[Conventional technology]

FIG. 8 shows an example of an optical arithmetic unit constituting a conventional feedback neural network.

In the figure, 81 is a light emitting diode array, 82 is a matrix optical mask, 83 is a light receiving element array, and 84 is a threshold processor.

Next, the operation of the above-mentioned conventional optical arithmetic device will be briefly explained. A fan-shaped light is irradiated from the light emitting diode array 81 onto the matrix optical mask 82 . Light emitting diode array 81
is a kind of vector v − (v t , V 2 , . . . , vj . . . , v
n). Further, the matrix optical mask 82 is divided into nXn elements, representing a type of matrix T=CT, . . . ) J in which each element has a different light transmittance. The j-th light emitting element of the light emitting diode 81 is used to illuminate the j-th row of the matrix optical mask 82, and the light transmitted through the i-th column of the optical mask 82 is received by the i-th light receiving element of the light receiving element array 83. You can perform vector and matrix multiplication such that .

By the way, neural computers generally store information on the strength of connections between neurons.

On the other hand, in the neural network composed of the optical arithmetic units described above, the light transmittance of each element of the matrix optical mask 82 is T1. It accumulates information. Its information storage rules are based on the Hopfield model. The light transmitted through the matrix optical mask 82 is received and outputted, and the vector is subjected to threshold processing by the threshold processor 84, and then the light emitting diode array 8
Give feedback to 1. In this way, for example, if three types of information corresponding to A, J, and E are stored in the matrix optical mask 82, incomplete information, such as A', is first input to the light emitting diode array as a vector. As the feedback is repeated, the output vector is given as representing the complete information A closest to the incomplete input information A'. In this way, it becomes possible to optically associatively recognize two-dimensional information such as characters and images.

[Problem to be solved by the invention]

Spatial light modulation elements have been widely used as matrix optical masks, which are the main components of optical processing devices. This spatial light modulation element has a function of modulating two-dimensional parallel optical information (for example, light intensity, phase, or polarization distribution) in real time using light modulation elements arranged in a matrix.

Therefore, if a spatial light modulation element is used, a dynamic neural network with a so-called learning effect can be constructed.

To date, we have developed liquid crystal materials, dielectric materials ('L IN b
Oa, PLZT, BSO single crystal, etc.), magnetic materials (YIG thin film), semiconductor materials (GaA
Various spatial light modulators using thin films have been used. However, these conventional spatial light modulation elements have a problem in that they do not have sufficient performance in terms of resolution, response speed, cost, ease of use, etc. In particular, in order to perform advanced and complex image recognition or image association, it is necessary to perform large-capacity vector and matrix operations at high speed, or it is difficult to obtain a high-density, large-capacity spatial light modulation element that satisfies such requirements. The problem was that it was not.

[Means to solve the problem]

In view of the above-mentioned problems of the conventional technology, the present invention provides high-density, high-density
An object of the present invention is to provide an integrated laminated optical operation element equipped with a large capacity and high speed spatial light modulation element.

In order to achieve this object, the optical arithmetic element according to the present invention has the basic structure shown in FIG. FIG. 1 is a schematic exploded cross-sectional view of an optical arithmetic element.

As shown in the figure, the optical arithmetic element has an X-axis two-dimensional L E'D array 1 . This LED array 1 has light emitting line parts arranged in rows, and constitutes a light emitting part that emits light in each row in accordance with each component value V of the vector inputted from the vector signal input means 8. Furthermore, it has a light valve 2. This light valve 2 has light modulation elements arranged in a matrix, and each component value T2 . An interspace light modulation section is configured to perform matrix modulation of the light emission according to the amount of light emitted. A Y-axis two-dimensional photodiode array 3 is arranged on the output side of the light valve 2. The photodiode array 3 has light-receiving line parts arranged in rows, receives light matrix-modulated by the light valve 2 in each row, and constitutes a light-receiving part for outputting a vector. The three components mentioned above are actually stacked one on top of the other to form a laminated structure. Then, the product operation of the input vector and the given matrix is optically executed. Furthermore, a comparator 4 is connected to the photodiode array 3, which performs threshold processing on each component value of the vector outputted from the photodiode array 3, and applies the resulting vector v,' again to the two-dimensional LED array. 1 constitutes a feedback means for inputting the input data. In this way, the illustrated optical arithmetic elements form a feedback neural network.

The light valve 2 is a high-density, high-speed, large-capacity spatial light modulation device, and is formed on a composite substrate consisting of a carrier layer and a semiconductor single crystal thin film layer formed thereon.

A plurality of light modulator electrodes are formed in a matrix on the semiconductor single crystal thin film layer. Further, a driving circuit including a switching element for driving a corresponding light modulation element electrode according to each component value of the matrix is integrally formed in the semiconductor single crystal thin film layer. A counter substrate is disposed opposite to the composite substrate with a predetermined gap therebetween, and this gap is filled with an electro-optic material layer that performs matrix light modulation under the control of individual light modulation element electrodes. Preferably, the carrier layer consists of quartz and the semiconductor single crystal thin film layer consists of silicon. This silicon semiconductor single crystal thin film layer has, for example, a crystal orientation of 100 > 0.
It has a uniformity in the range of 0±1.0, and its single crystal lattice defect density is 500 defects/ant or less. By using such a high quality silicon semiconductor single crystal thin film layer, VLSI manufacturing technology can be applied as is, and 1000 rows x 10
It becomes possible to form the light modulation elements in the OO array with high precision.

Note that twisted oriented nematic liquid crystal can be used as the electro-optic material layer. Furthermore, in order to improve the responsiveness of the light valve, a ferroelectric liquid crystal having a bistable state can be used as the electro-optic material layer.

Returning to FIG. 1 again, the remaining accessory components will be explained.

A pair of polarizing plates A and B are arranged before and after the light valve 2. Since the light valve 2 normally performs polarization modulation, in order to convert this into a change in light intensity, the polarizing plate A functions as a polarizer and the polarizing plate B functions as an analyzer. A cylindrical lens array 5 having a generatrix in the X direction, that is, the row direction, is arranged between the LED array 1 and the light valve 2. This cylindrical lens array 5 transmits light emitted row by row from the LED array 1 to a light valve 2.
It has a function to focus light on the corresponding row.

Further, a cylindrical lens array 6 having a generating line in the Y direction, that is, the column direction, is inserted between the light valve 2 and the Y-axis two-dimensional photodiode array 3. This lens array 6 is for condensing the light that has passed through the light valve 2 onto the light receiving line portion of the photodiode array 3 corresponding to each column. The cylindrical lens array can be provided as a separate component, but preferably, for example, the cylindrical lens array 5 on the input side may be formed using thermal diffusion in the carrier layer constituting the composite substrate of the light valve 2.

In the basic configuration described above, the light emitting section is composed of a two-dimensional LED array. However, a surface emitting device can also be used. In this case, an additional spatial light modulator having a structure similar to that of the light valve 2 is arranged on the light emitting surface of the surface emitting element, and the light emitting surface is substantially divided into rows, and each component of the input vector is The light emission intensity for each row may be controlled according to the value.

[Action of the invention]

As shown in FIG. 1, an optical operation element composed of a two-dimensional LED array 1, a light valve 2, a photodiode array 3, a comparator 4, etc. constitutes a feedback type neural network. The excited state of the neuron corresponds to the blinking state of each light emitting element of the LED array arranged in rows.

Each light-emitting element emits light according to the component value V of the input vector, and only the j-row component of the light valve 2 corresponding to the matrix T is uniformly irradiated via the cylindrical lens array 5 having a row-direction generating line. do. Each component value T1 of matrix T. If the magnitude of, is given as the light transmittance of each modulation element of the light valve J2, its output light intensity is proportional to T,′,v,. Next, this
The output light from the light valve 2 of J is focused on all column components or corresponding light receiving elements of the photodiode array 3 by a cylindrical lens array 6 having a generatrix in the column direction. Therefore, the first photodetector output v,' is matrix-
A vector product is obtained. In this way, since calculations are performed completely optically, extremely high-speed processing can be performed in parallel. The output signal photoelectrically converted by the photodiode array 3 is subjected to threshold processing by the comparator 4, and then output to the LED array 1.
will be given feedback. Through this iterative operation, the most similar information is selected from among the plurality of complete information stored in the matrix of the light valve 2 for the input vector containing incomplete information, and the complete (oo) total output V, is obtained. . At this time, by using the light valve 2 using a single crystal thin film layer, a large capacity matrix of about 500 x 500 to 100 x 1000 can be realized. As a result, the number of input characters that can be associated can be increased from 30,000 to 100,000, and handwritten kanji input can also be recognized.

〔Example〕

Preferred embodiments of the present invention will be described in detail below with reference to the drawings. FIG. 2 is a schematic perspective view showing an embodiment of the integrated layered optical arithmetic element according to the present invention. As shown in the figure, the optical arithmetic element is mounted on a heat sink 21. A gallium arsenide substrate 22 is placed on the heat sink 21 . In this embodiment, a gallium arsenide material is used, but a gallium phosphide material may also be used. An X-axis two-dimensional LED array 23 is formed on the surface of the gallium arsenide substrate 22. This LED array 23 consists of a plurality of light emitting line elements arranged in the X/J direction or in the row direction. These LED arrays 23
An LED electrode terminal 24 for supplying current to the LED electrode terminal 24 is also formed on the same surface. Polarizing plates A are stacked with an air gap 25 in between so as to cover the LED array 23.

A light valve is arranged on the polarizing plate A. This light valve has a composite substrate in which a single crystal silicon layer 27 is formed on a quartz substrate 26. In the single crystal silicon layer 27, a plurality of light modulation element electrodes formed in a matrix and a driving circuit including a switching element such as a transistor for selectively exciting each electrode are arranged in an integrated manner. In addition, an LED driver 28 for driving the two-dimensional LED array 23 is also formed in the single crystal silicon layer 27. Since ordinary VLSI manufacturing technology can be applied to the single crystal silicon layer 27, it is also possible to form IC memories, photodiode drive circuits, comparators for threshold processing, and the like. Further, on the back side of the quartz substrate 26, a cylindrical lens array 29 having an X-direction generatrix is formed by thermal diffusion technology in order to prevent optical crosstalk.

The individual cylindrical lens elements constituting the cylindrical lens array 29 are arranged to correspond to the individual light emitting line elements constituting the LED array 23. Spacer 30 is placed on the composite board.
Opposing glass substrates 31 are arranged to face each other with a predetermined gap therebetween. This gap is filled with a liquid crystal layer 32 as an electro-optical material layer that performs light modulation. liquid crystal layer 32
is sandwiched between a pair of alignment films 33 and 34 from above and below, and has a predetermined liquid crystal molecule alignment. Opposing glass substrate 3
A common transparent electrode 35 is formed on the inner surface of the liquid crystal layer 32, and the alignment state of the liquid crystal layer 32 is adjusted according to the voltage value applied between the common transparent electrode 35 and each light modulator electrode formed on the single crystal silicon layer 27. changes. An X-direction two-dimensional optical mask 36 is formed on the front surface side of the opposing glass substrate 31, and prevents the light output from the light valve from causing crosstalk between each row. A polarizing plate B is attached on the X-direction two-dimensional optical mask 36. Another quartz substrate 37 is arranged on the polarizing plate B. A cylindrical lens array 38 having a generatrix in the Y direction and the row direction is formed on the back side of the quartz substrate 37 by thermal diffusion. Individual cylindrical lens elements making up the cylindrical lens array 38 are arranged to correspond to each row of light valves. A Y-axis two-dimensional photodiode array 39 is formed on the front side of the quartz substrate 37 using thin film technology. Each light receiving line element constituting the photodiode array 39 is formed to correspond to each row of light valves. The photodiode array 39 is connected to a photodiode amplifier 40 formed on the same thin film. This photodiode amplifier 40 is connected to a comparator formed in the single crystal silicon layer 27 that performs threshold processing.

As described above, the LED array 23, the light valve, and the photodiode array 39 are each integrated and stacked on each other to form an optical operation chip or an optical neuron chip. In particular, the light valve is equipped with a single-crystal silicon layer 27, and by integrating a matrix of light modulation element electrodes, switching elements, etc. in this layer at a density on the submicron or micron order, a large capacity matrix can be obtained. can. In addition, IC memory, LED array drive]
6. Since the dynamic circuit, photodiode array drive circuit, comparator, etc. can be formed simultaneously using VLSI technology, an extremely compact configuration can be achieved. or,
Further hybridization becomes possible by diffusing and forming the cylindrical lens array 29 or 38 within the quartz substrate.

Next, FIG. 3 shows another embodiment of the optical arithmetic element according to the present invention. In this embodiment, an EL clamp, which is a surface emitting element, is used in place of the X-axis two-dimensional LED array. This EL clamp 1 is normally used as a backlight for a liquid crystal display element, and can be driven with lower power consumption than a two-dimensional LED array. E, which is a surface emitting element
An additional light valve is arranged above the L-clamp 1,
Substantially, the light emitting surface is divided into rows, and the light emitting intensity for each row is controlled according to each component value of the input vector. This additional light valve has essentially the same structure as the matrix light valve shown in FIG. That is, a quartz substrate 43 is placed on an additional polarizing plate 42 for imparting polarization to the light emitted from the EL clamp 1. A single crystal silicon layer 44 is formed on the surface of the quartz substrate 43, forming a composite substrate. In this single crystal silicon layer 44, a light modulation element electrode for controlling the emission intensity for each row according to each component value of the input vector, a switching circuit for driving this electrode, etc. are formed.

On top of the composite substrate, a glass substrate 46 is placed oppositely with a spacer 45 in between, with a predetermined gap being maintained. A liquid crystal layer 47 whose transmittance changes depending on the voltage is filled within a predetermined space. The liquid crystal layer 47 is sandwiched between a pair of alignment films 48 and has a predetermined alignment of liquid crystal molecules. A transparent electrode 49 is disposed on the inner surface of the glass substrate 46, and is connected to the individual light modulator electrodes formed on the single crystal silicon layer 44 to transmit data to the liquid crystal layer 47 according to each component value of the input vector. A voltage of a certain magnitude is applied to locally change the transmittance of the liquid crystal layer 47 to control the emission intensity for each row. In this way, the quartz substrate 43, the single crystal silicon layer 44, the liquid crystal layer 47,
An additional light valve is constituted by a glass substrate 46 or the like. Since the portions superimposed on this additional light valve are the same as the embodiment shown in FIG. 2, corresponding components will be designated by the same numerals and their description will be omitted.

FIG. 4 is a schematic exploded perspective view showing the detailed structure of a light valve that forms the main part of the optical arithmetic element according to the present invention. As shown in the figure, the light valve includes a composite substrate (28, 27), a glass substrate 31 disposed opposite to the composite substrate, and an electro-optic material layer such as a liquid crystal layer disposed between the composite substrate and the glass substrate 31. It is composed of a layer 32.

A modulation element electrode or drive electrode 51 defining a light modulation element and a drive circuit for exciting the drive electrode 51 in accordance with a predetermined signal, that is, each component value of a given matrix are formed on the composite substrate.

The composite substrate has a two-layer structure consisting of a quartz glass substrate 26 and a single crystal silicon semiconductor film layer 27. In addition, a polarizing plate A is bonded to the back side of the quartz glass substrate 26.

Furthermore, a cylindrical lens array 29 is formed within the back surface of the substrate 26. The drive circuit is composed of an integrated circuit formed in this single crystal silicon semiconductor film layer 27. This integrated circuit includes a plurality of field effect insulated gate transistors 52 arranged in a matrix. The source electrode of the transistor 52 is connected to the corresponding drive electrode 51 , the gate electrode is similarly connected to the scanning line 53 , and the drain electrode is similarly connected to the signal line 54 . The integrated circuit further includes an X driver 55 and is connected to a column-shaped signal line 54. Furthermore, it includes a Y driver 56 and is connected to the row scanning lines 53 .

Further, the counter glass substrate 31 has a polarizing plate B adhered to its outer surface, and a counter electrode or common transparent electrode 35 formed on its inner surface. In addition, the inner surface 2 of the composite substrate provided with the drive electrode 51 and the drive circuit that drives it.
7 and the inner surface of the counter electrode 35 are each provided with an alignment film 2.
7a and 35a are formed.

FIG. 5 is a schematic perspective view showing one light modulator of the light valve shown in FIG. 4 cut away, and FIG. 5(A) shows the case where the light modulator is in a non-selected state. , FIG. 5(B) shows the case where the light modulator is in the selected state.

In this embodiment, a nematic liquid crystal material is used as the liquid crystal layer 32. Nematic liquid crystal molecules have the property that their long axes are easily aligned.

The orientation of liquid crystal molecules is determined by the quartz substrate 26 and the opposing glass substrate 31.
It is obtained by rubbing the inner surface of the Since the rubbing direction differs by 90 degrees between the upper and lower boards,
As shown in the figure, the liquid crystal molecules also rotate 90 degrees following this.

As a result, the polarization axis of the light passing through this liquid crystal layer 32 is 90°.
It will rotate. On the other hand, as shown in FIG. 5(B), the drive electrode 51 formed on the surface of the quartz substrate 26 and the counter electrode 3 formed on the inner surface of the counter glass substrate 31
When an electric field is applied between 5 and 5, the liquid crystal molecules are aligned in the direction of the electric field, that is, perpendicular to the substrate, and the optical rotation is lost. This transition is optically detected by a pair of polarizing plates A and B placed above and below the liquid crystal layer. That is, light passing through the liquid crystal layer is transmitted or blocked depending on whether or not a voltage is applied. In this way, the light valve according to the present invention has a light modulation function for each light modulation element.

Next, the operation of the light valve described above will be explained in detail with reference to FIGS. 4 and 5. The gate electrode of each transistor element 52 is connected to a scanning line 53, and a scanning signal is applied by a Y driver 56 to control conduction and cut-off of each transistor element 52 line sequentially. A signal representing a matrix component outputted from the X driver 55 is connected to a signal line 54.
is applied to the selected transistor 51 which is in a conductive state. The applied signal is transmitted to the corresponding drive electrode 51, excites the drive electrode, acts on the liquid crystal layer 32, and makes its transmittance substantially 100%. On the other hand, when not selected, the transistor element 52 becomes non-conductive and maintains the signal written to the drive electrode as a charge. The matrix given in this way is expressed as the transmittance of each modulator.
Written on the light valve. Note that the liquid crystal layer 32 has a high specific resistance and normally operates as a capacitor.

An on/off current ratio is used to express the switching performance of these drive transistor elements 52.

In the present invention, since the drive transistor element is formed in the single crystal silicon semiconductor film layer 27 having extremely high charge mobility, an on/off ratio of 6 digits or more can be ensured. Therefore, it is possible to obtain an active matrix type light valve with extremely high-speed signal response, which is optimal as a spatial modulation element for a dynamic optical neural network. Further, by utilizing the high mobility characteristics of a single crystal thin film, it becomes possible to simultaneously form a peripheral driver circuit on the same silicon single crystal semiconductor thin film.

Next, an example in which a ferroelectric liquid crystal is used as the electro-optic material layer is shown in FIGS. 6(A) and 6(B). These figures are schematic top views of one modulation element of a light valve, showing the molecular arrangement of the ferroelectric liquid crystal. When a ferroelectric liquid crystal such as a chiral smectic liquid crystal is filled into a gap with a thickness of about 111rn, it is oriented in a bistable state. FIG. 6(8) shows one bistable state, and FIG. 6(B) shows the other stable state. These bistable states can be switched rapidly between each other depending on the direction of the applied voltage. By this switching, the transmittance of light passing through the liquid crystal layer can be controlled. A light valve that utilizes switching of the bistable state of such a ferroelectric liquid crystal has superior responsiveness compared to the light valve that uses the above-mentioned twisted nematic liquid crystal, and can perform faster dynamic calculation processing. In particular, it is suitable for expressing a binary matrix in which each component value can take either 0 or 1 using a bistable state.

It should be noted that it is better to use silicon monoxide with a predetermined thickness at an angle of 75 degrees to 85 degrees from the normal direction of the substrate as the alignment film used in the above two examples. This is preferable because it causes less light scattering and does not damage the driving integrated circuit of the light valve formed on the surface.

Finally, a method for manufacturing a semiconductor chip-type light valve substrate in which a modulator electrode and a drive circuit are integrated will be described in detail with reference to FIG. First, in the step shown in FIG. 7(A),
A quartz glass substrate 71 is prepared.

Then, a cylindrical lens region 72 constituting a cylindrical lens array is formed on the back side of this substrate 71 by thermal diffusion. This thermal diffusion is carried out, for example, by forming a titanium vacuum-deposited film in a stripe shape and then performing a heat treatment at a temperature of 1000° C. to 1100° C. for 6 hours to 10 hours.

Next, in the step shown in FIG. 7(B), a single crystal conductor substrate 73 is attached to the quartz glass substrate 71. It is preferable to use a high-quality silicon wafer used for LSI manufacturing as the single crystal silicon semiconductor substrate 73, and its crystal orientation has uniformity in the range of <1'00>0.0±1.0. Single crystal lattice defect density is 500/Ca
It is as follows. First, the surface of the quartz glass substrate 71 on which the lens region 72 is formed and the surface of the single crystal silicon semiconductor substrate 73 are precisely smoothed.

Subsequently, the side substrates are thermocompression bonded by overlapping and heating the smoothed surfaces. By this thermocompression bonding process, the side substrates 71 and 73 are firmly fixed to each other.

In the step shown in FIG. 7(C), the surface of the single crystal silicon semiconductor substrate 73 is polished. As a result, a 111-crystalline silicon semiconductor film layer 74 polished to a desired thickness is formed on the surface of the quartz glass substrate 71.

A composite substrate having a two-layer structure consisting of a carrier layer made of a quartz glass substrate and a single crystal silicon semiconductor film layer is obtained. Note that in order to reduce the thickness of the 11-crystalline silicon semiconductor substrate 73, an etching process may be used instead of the polishing process. Since the quality of the silicon wafer 73 is substantially preserved in the single-crystal silicon semiconductor film layer 74 obtained in this way, it is possible to obtain a substrate material that is extremely excellent in terms of uniformity of crystal orientation and lattice defect density. . On the other hand, single-crystal thin films obtained by recrystallizing deposited amorphous or polycrystalline silicon thin films, as has been done conventionally, have many lattice defects and uneven crystal orientation, making them suitable for LSI.
Not suitable for manufacturing.

In the step shown in FIG. 7(D), the surface of the single crystal silicon semiconductor film layer 74 is thermally oxidized to form a silicon oxide film 75 on the entire surface. A silicon nitride film 76 is deposited thereon using chemical vapor deposition. Furthermore, a resist 77 is coated. The resist 77 is patterned by photolithography and removed leaving only the element region 78. In this state, an etching process is performed to remove portions of the silicon oxide film 75 and silicon nitride film 76 that are not covered by the resist 77.

In the step shown in FIG. 7 (IE), after removing the resist, thermal oxidation treatment is performed on the single crystal silicon semiconductor film layer 74 using the silicon oxide film 75 and silicon nitride film 76 covering the finished region 78 as a mask to perform field oxidation. A film 79 is formed. Single crystal silicon film layer 74 is left in a region surrounded by field oxide film 79 to form element region 78 . In this state, the silicon oxide film and silicon nitride film used as a mask have been removed.

In the step shown in FIG. 7(P), thermal oxidation treatment is performed again to form a gate oxide film 80 on the surface of the single crystal silicon film layer 74.

In the step shown in FIG. 7(G), a polycrystalline silicon film is deposited by chemical vapor deposition. This polycrystalline silicon film is selectively etched using a resist 81 patterned into a predetermined shape to form a gate electrode 82 made of a polycrystalline silicon film on the gate oxide film 80.

In the step shown in FIG. 7(]I), after removing the resist I-81, impurity arsenic ions are implanted through the gate oxide film 80 using the gate electrode 82 as a mask, and the drain region is implanted into the silicon t11-crystal film. 83 and Soth area 8
form 4. As a result, a transistor channel forming region 85 in which impurity arsenic is not implanted between the drain region 83 and the source region 84 below the gate electrode 82
will be provided.

Finally, in the second step shown in FIG. 7 (+), a part of the gate oxide film 80 above the drain region 83 is removed to form a contact hole to which a drain electrode 86 is connected. Similarly, a portion of the gate oxide film 80 above the source region 84 is removed to form a contact hole, and a drive electrode 87 is formed to fill this portion. Drive electrode 8
7 or the light modulator electrode is composed of a transparent electrode made of ITO or the like.

In addition, the field oxide film 79 disposed below the light modulator electrode 87 is also transparent, and the quartz glass substrate 71 disposed below it is also transparent. Therefore, the three-layer structure consisting of the drive electrode 87, the field oxide film 79, and the quartz glass substrate 71 is optically transparent, and a transmission type light valve can be obtained.

Note that the cylindrical lens region 72 is not limited to the process order shown in FIG. 7, but is preferably formed in the most appropriate process.

As mentioned above, in the manufacturing method shown in FIG.
By performing high-resolution photolithography, ion implantation, etc., it is possible to form a field effect type insulated gate transistor having a size on the micron order or submicron order. Since the silicon single crystal film used is of extremely high quality, the electrical properties of the resulting insulated gate transistor are also excellent. At the same time, the light modulator electrode can also be formed with dimensions on the micron order using miniaturization technology, so it is possible to manufacture a semiconductor integrated circuit chip substrate for an active matrix liquid crystal light valve having a high density and fine structure.

Finally, a method for learning two-dimensional information such as images using the optical arithmetic element of the present invention will be explained.

In a network composed of N neurons, the number of two-dimensional information to be stored is Ml, and the state of the neuron for the m-th two-dimensional information is a binary (O or 1) vector ν(m), , (m )(1'1 (m) ... (m)), it is possible to learn the input vector by calculating the components of the matrix "2 . . . I"N. Are known. The present invention is a dynamic neural network in which the components of the matrix are variable. Therefore, all vectors to be recognized are T2. Needless to say, it is possible to perform pattern recognition or vector matrix calculations after calculating T1 for any new input vector during the recognition process. can be learned by calculating and immediately used as a newly learned matrix component. This is one example of how the present invention is extremely flexible compared to conventional optical arithmetic elements using fixed optical masks.

〔Effect of the invention〕

As described above, according to the present invention, it is possible to obtain an integrated laminated optical operation element or an optical operation chip that incorporates an ultra-high density, ultra-high speed, and ultra-large capacity spatial light modulation element. Therefore, as many as 106 light modulation elements can be integrated in such a spatial light modulation element, which has the effect of enabling high-speed arithmetic processing of large-scale two-dimensional information. In addition, by combining the additional spatial light modulation element and the planar light emitting element of the EL clamp, it is possible to construct an inexpensive, practical, and low power consumption optical calculation chip if used in place of the semiconductor LED array. There is. Furthermore, in addition to the light modulation element electrode and the switching transistor, a peripheral drive circuit,
Since logic circuits or memories can be formed at the same time, an extremely small optical arithmetic chip can be constructed. Furthermore, since a cylindrical lens array having a light condensing function can be formed in the quartz glass constituting the composite substrate using thermal diffusion technology, a more advanced hybridization of the optical processing chip can be achieved. In addition, by using a ferroelectric liquid crystal having a bistable state as the electro-optical material layer of the spatial light modulator, there is an effect that arithmetic processing using binary 7-trix can be performed at extremely high speed.

[Brief explanation of the drawing]

FIG. 1 is a schematic exploded side view showing the basic configuration of an integrated laminated optical processing element according to the present invention, and FIG. 2 is a schematic perspective view showing an embodiment of the optical processing element according to the invention. FIG. 3 is a schematic side view showing another embodiment of the optical arithmetic element, FIG. 4 is a schematic enlarged perspective view showing the detailed structure of a spatial light modulation element or light valve built into the optical arithmetic element, and FIG. (A) and FIG. 5(B) are schematic perspective views for explaining the operation of individual light modulators included in the spatial light modulator, and FIGS. 6(A) and 6(B) are spatial light modulators. A schematic plan view for explaining the operation when a bistable ferroelectric liquid crystal is used as the electro-optic substance of the light modulation element, and FIG. 7(A) or 1) shows the schematic plan view of the spatial light modulation element. FIG. 8 is a process diagram showing a method of manufacturing a semiconductor chip that can be used as a substrate, and FIG. 8 is a schematic diagram showing a conventional optical arithmetic element. 1...X-axis two-dimensional LED array 2...Light valve 3...X-axis two-dimensional photodiode array 4...Threshold processing comparator 5...Cylindrical lens array having an X-direction generatrix 6...・
Cylindrical lens array 7 having a Y-direction generatrix...matrix signal input means 8...vector signal input means

Claims (1)

[Scope of Claims] 1. A light emitting unit having a light emitting line section arranged in rows and emitting light in each row according to each component value of an input vector, and a light modulating element arranged in a matrix. It has a spatial light modulation section for matrix modulating the light emission according to each corresponding component value of a given matrix, and a light reception line section arranged in rows, and receives the modulated light in each row. and a light-receiving section for outputting a vector having a component value corresponding to each received light intensity, in an optical operation element that performs a product operation of an input vector and a given matrix, the spatial light modulation section comprises a composite substrate comprising a carrier layer and a semiconductor single-crystal thin film layer formed thereon, a plurality of light modulator electrodes formed in a matrix on the semiconductor single-crystal thin film layer, and the semiconductor single-crystal thin film. a drive circuit that is integrally formed in a layer and drives a corresponding light modulation element electrode according to each component value of the matrix; and a counter substrate that is disposed opposite to the composite substrate with a predetermined gap therebetween. an electro-optical material layer disposed in the gap and controlled by individual light modulator electrodes to perform matrix light modulation. 2. The optical arithmetic element according to claim 1, wherein the carrier layer is made of quartz and the semiconductor single crystal thin film layer is made of silicon. 3. The silicon semiconductor single crystal thin film layer has a crystal orientation of <
3. The optical arithmetic element according to claim 2, having uniformity in the range of 100>0.0±1.0, and having a single crystal lattice defect density of 500 defects/cm^2 or less. 4. A cylindrical lens array arranged in rows for condensing the light emitted row by row onto a corresponding row of the spatial light modulator is formed on the carrier layer. 1
The optical arithmetic element described in . 5. The optical operation element according to claim 1, wherein the electro-optic material layer is made of twisted nematic liquid crystal. 6. The optical operation element according to claim 1, wherein the electro-optic material layer is made of ferroelectric liquid crystal having a bistable state. 7. The light-emitting section includes a surface-emitting element, and is arranged on the light-emitting surface of the surface-emitting element, substantially dividing the light-emitting surface into rows, and emitting light intensity for each row according to each component value of the input vector. The optical arithmetic element according to claim 1, further comprising an additional spatial light modulator for control. 8. Feedback means is provided for thresholding each component of the vector output from the light receiving section and inputting the resulting vector to the light emitting section again, forming a feedback neural network. The optical arithmetic element according to claim 1.