WO2022157853A1 - Photonics/electronics integrated computer - Google Patents

Photonics/electronics integrated computer Download PDF

Info

Publication number
WO2022157853A1
WO2022157853A1 PCT/JP2021/001839 JP2021001839W WO2022157853A1 WO 2022157853 A1 WO2022157853 A1 WO 2022157853A1 JP 2021001839 W JP2021001839 W JP 2021001839W WO 2022157853 A1 WO2022157853 A1 WO 2022157853A1
Authority
WO
WIPO (PCT)
Prior art keywords
circuit
output
optical
decision
pulse
Prior art date
Application number
PCT/JP2021/001839
Other languages
French (fr)
Japanese (ja)
Inventor
イブラヒム サラ
俊和 橋本
Original Assignee
日本電信電話株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電信電話株式会社 filed Critical 日本電信電話株式会社
Priority to JP2022576270A priority Critical patent/JP7495646B2/en
Priority to PCT/JP2021/001839 priority patent/WO2022157853A1/en
Publication of WO2022157853A1 publication Critical patent/WO2022157853A1/en

Links

Images

Classifications

    • HELECTRICITY
    • H03ELECTRONIC CIRCUITRY
    • H03MCODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M7/00Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
    • HELECTRICITY
    • H03ELECTRONIC CIRCUITRY
    • H03MCODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M9/00Parallel/series conversion or vice versa

Definitions

  • the present invention relates to an optoelectronic convergence computer, and more particularly to an optoelectronic convergence computer provided with a computing platform that combines optical pulses and electrical pulses.
  • a chip for digital signal processing contains electronic circuits densely integrated with a large number of logic gates.
  • these electronic circuits have been implemented with complementary metal oxide semiconductor (CMOS) transistors.
  • CMOS complementary metal oxide semiconductor
  • the miniaturization of CMOS transistors has progressed, electronic circuits with high integration densities have been realized, the clock speed has been increased, and the function and performance of signal processing have been continuously improved.
  • current CMOS transistors have been miniaturized to dimensions of several nanometers, and further miniaturization has become difficult due to various technical issues and high costs.
  • the object of the present invention is to provide an opto-electronic convergence computer equipped with a computing platform capable of high-speed operation by combining optical pulses and electrical pulses.
  • an opto-electronic integrated computer which is a recognition circuit for recognizing an execution instruction including an operator and an operand output from a controller, comprising: a serial-to-parallel converter outputting a light pulse that controls a bit-by-bit corresponding decision stage, and said light pulse selects the output of said decision stage to drive a control signal from an output port uniquely corresponding to said execution instruction. It is characterized by comprising a recognition circuit including a decision circuit for outputting, and a memory for outputting a pre-stored calculation result according to the control signal.
  • FIG. 1 is a diagram showing signal processing in an ultra-high-speed pattern recognition circuit according to one embodiment of the present invention
  • FIG. 2 is a diagram showing the configuration of an ultra-high-speed pattern recognition circuit capable of processing 4-bit words according to this embodiment
  • FIG. 3 is a diagram explaining how to correspond an input word represented by 4 bits to an output represented by a decimal number
  • FIG. 4 is a diagram showing a processing sequence in the ultra-high-speed pattern recognition circuit according to this embodiment
  • FIG. 5A is a diagram showing an example of an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to the present embodiment
  • FIG. 5B is a diagram showing another example of an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to the present embodiment
  • FIG. 6A is a diagram showing an opto-electric hybrid logic circuit of a decision unit according to the present embodiment
  • FIG. 6B is a diagram showing a truth table representing the input/output relationship of the opto-electric hybrid logic circuit
  • FIG. 7 is a diagram for explaining the principle of a computing platform according to one embodiment of the present invention
  • FIG. 8 is a diagram showing the configuration of the computing platform of this embodiment
  • FIG. 9 is a diagram showing a schematic configuration of the optical memory of the computing platform of this embodiment
  • FIG. 10 is a diagram showing another configuration example of the optical memory of this embodiment.
  • Optical pulses can be generated with low jitter and can propagate long distances with low loss and dispersion. Also, by adjusting the length of the optical waveguide through which the optical pulse propagates, the timing of the optical pulse can be controlled accurately. For example, a 20 ⁇ m optical fiber produces a delay of approximately 0.1 ps. These characteristics make it suitable for high-speed digital signal processing, but unlike logic gates made up of CMOS transistors, it is difficult to integrate optical logic gates on a large scale. Therefore, we will construct an optoelectronic convergence computing platform that combines optical and electrical pulses.
  • FIG. 1 shows signal processing in an ultra-high speed pattern recognition circuit according to one embodiment of the present invention.
  • the recognition circuit 100 has one serial port to which a high-speed bit signal is input, and 2 N output ports that recognize an N-bit input word and uniquely correspond to 2 N bit combinations.
  • the output ports of the recognition circuit 100 are connected to 2 N input ports of the electrical circuit 130 (see, for example, Patent Document 1).
  • Recognition circuit 100 generates output signals at spatially separated output ports corresponding to the bit combinations based on the bit combinations of the input word.
  • the duration of the signal output from this output port corresponds to the duration of one word (T word ), i.e. the duration of all the bit signals, so it can be adapted to the speed of slow electrical circuits. is sufficiently long.
  • T word duration of one word
  • the output port to which the generated signal is output contains the collective information of all bit signals, that is, the information of the bit combination, it is possible to consume less clock cycles than in the past, and the processing in the electric circuit is reduced. can be simplified.
  • each word When a number of consecutive N-bit words are input at arbitrary time intervals (T next ) into M words that need to be logically operated on, each word is spatially distinct according to its bit combination. It is output to the output port at time intervals (T next ).
  • T next time intervals
  • a logic circuit capable of processing these M consecutive slow signals performs the operation.
  • Outputs corresponding to each of the M words are sequentially input to electrical circuit 130 .
  • the first input is the first processed result, but the next input is processed using the previous result and updated until all operations are completed. This establishes a word-by-word processing method that can reduce the processing time of ultra-high-speed bits with a low-speed electric circuit.
  • FIG. 2 shows the configuration of an ultra-high-speed pattern recognition circuit capable of processing 4-bit words according to this embodiment.
  • Recognition circuit 100 consists of two major functional blocks, serial-to-parallel converter 110 and decision circuit 120 .
  • the output of each conversion channel of serial-to-parallel converter 110 controls a predetermined decision stage S of decision circuit 120 .
  • the most significant bit determines whether the final output is less than 8 or greater than 8 depending on whether the state is High level or Low level. Therefore, if the value of the most significant bit is known, it is possible to halve the possible values of the final output. If the state of the next higher-order bit is known, the remaining value candidates can also be halved, that is, the final output value candidates can be narrowed down to 1/4. By repeating this procedure down to the least significant bit, the possibility of improper output is continuously eliminated, and the word is converted to the correct output, that is, only the output from the output port corresponding to the bit combination of the word is set to a high level. state can be made. Therefore, the decision circuit 120 is configured as follows.
  • the decision stage S1 corresponding to the most significant bit includes one decision unit U 1,1
  • the decision stage S2 corresponding to the bit below the most significant bit comprises two decision units U 2,1 , U 2,2 .
  • the decision stage S3 corresponding to the one lower bit includes four decision units U 3,1 to U 3,4 .
  • the decision stage S4 corresponding to the least significant bit comprises eight decision units U 4,1 to U 4,8 .
  • the two output ports of the decision unit U 1,1 in the decision stage S1 corresponding to the most significant bit are connected to the decision unit U 2,1 of the decision stage S2 corresponding to the second most significant bit, the other is connected to the decision unit U 2,2 .
  • the four output ports of decision units U 2,1 , U 2,2 of decision stage S2 are connected to the four decision units U 3,1 to U 3, of decision stage S3 corresponding to the third most significant bit.
  • the eight output ports of the four decision units U 3,1 to U 3,4 of the decision stage S3 are connected to the eight decision units U 4,1 to U of the decision stage S4 corresponding to the least significant bits. Connected to 4,8 .
  • the decision unit U 1,1 of the first decision stage S 1 selects one of the two output ports based on the control signal C 1 generated by the conversion channel that converts the most significant bit signal of the serial-parallel converter 110 .
  • One is set to High level.
  • the decision unit U 1,1 When one of the outputs of the decision unit U 1,1 is brought to a high level, it activates only one of the two decision units U 2,1 , U 2,2 of the second decision stage S2.
  • the activated decision unit U 2,1 or U 2,2 makes two decisions based on the control signal C 2 generated in the conversion channel converting the second most significant bit signal of the serial-to-parallel converter 110 .
  • One of certain output ports is set to High level.
  • a high level signal from the second decision stage S2 activates one of the four decision units U 3,1 to U 3,4 of the third decision stage S3.
  • One of the activated decision units U 3,1 to U 3,4 then receives the control signal C 3 generated in the conversion channel converting the third most significant bit signal of the serial-to-parallel converter 110 .
  • one of the two output ports is set to High level.
  • Any one of the eight decision units U 4,1 to U 4,8 in the fourth decision stage S4 is activated by the High level signal from the third decision stage S3, and the serial-parallel converter 110 One of the two output ports is set to High level based on the control signal C4 generated by the conversion channel that converts the least significant bit signal of .
  • an input word represented by 4 bits can be associated with an output represented by a decimal number, and the output of one of the 16 output ports can be set to a high level. For example, if the 4-bit word "1101" is input to the recognition circuit 100, different binary combinations may correspond to "11", one of the decimal integers 0-15.
  • the recognition circuit 100 is not limited to the processing of the 4-bit words described above. By repeatedly executing the above procedure for N-bit words having arbitrary bit combinations, the output corresponding to the bit combinations for each word is obtained. Only the output of the port can be brought to a high level state.
  • the serial-parallel converter 110 has N conversion channels corresponding to each bit constituting a word
  • the decision circuit 120 has N stages corresponding to the N conversion channels of the serial-parallel converter 110. It is assumed that there are decision stages S 1 to S N .
  • Each decision unit U has two output ports, each output port is connected one-to-one to a different decision unit U in the next lower decision stage S, one of which U only activate. At each decision stage S, only one decision unit U is activated at a time by a high level output from the decision stage S above it.
  • the decision units U belonging to the same decision stage S are connected in parallel to the same conversion channel of the serial-to-parallel converter 110, and the output of the activated decision unit U is the decision stage S to which the decision unit U belongs. is controlled by a control signal C generated in the conversion channel of the serial-to-parallel converter corresponding to .
  • a control signal C generated in the conversion channel of the serial-to-parallel converter corresponding to .
  • FIG. 4 shows a processing sequence in the ultra-high speed pattern recognition circuit according to this embodiment.
  • Control signal C1 determines only the output of unit U 1,1 and converts signal L 1,1 to a high level since the most significant bit is a high level in this example.
  • the signal L 1,2 remains at a steady-state low level.
  • the circuit sets the duration of signal L 1,1 to 4T (T: clock cycle time) to allow for repeated operations, and when a new word arrives after time 4T, unit U 1,1 can again be freely determined.
  • T clock cycle time
  • any variation in output onset at each stage affects the duration of the signal at the final output of the circuit.
  • a signal generated in each decision stage S is used to control only one decision unit U among the decision units U in the next decision stage. That is, the signal needs to move very few transistors without creating an electrical load that prevents fast operation. Also, to allow processing of high speed electrical signals in such an arrangement, the integrated circuit must be designed with dimensions to accommodate the speed of the signals being processed.
  • each control signal C n issued by the serial-to-parallel converter 110 to a particular decision stage S n must control all decision units belonging to that decision stage.
  • the control signals must be routed through numerous transistors with large capacitive loads, hindering rapid operation. Also, if the capacitive load is large, the rise time will be longer than necessary, extending the duration of the signal. Therefore, an opto-electric hybrid circuit combining an optical pulse and an electric pulse, which will be described below, is applied.
  • FIG. 5A shows an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to this embodiment.
  • the output of each conversion channel 111 of the serial-to-parallel converter 110 is connected to an optical waveguide 121 located adjacent to the row of decision units involved in the decision stage S.
  • the decision unit U includes an optical splitter 122 for drawing light pulses from the optical waveguide 121 into the decision unit U according to the signal L n-1,j for activating the decision unit U from the previous decision stage S; and an opto-electric hybrid logic circuit 123 which determines whether to output an electric pulse from one of the two output ports according to two input signals of an electric pulse and an optical pulse.
  • a control signal C n which is a light pulse is generated from the conversion channel 111 of the serial-to-parallel converter 110 towards the decision unit U of the decision stage S, while the converted If the bit signal is low, no light pulse is generated.
  • the signal L n-1,j is used to activate the decision unit U n,i .
  • the control signals C n which are light pulses are activated.
  • the signal L n-1,i is used to deflect to the determined decision unit U n,i .
  • the signal L n-1,i is split, the signal L n-1,i controls the optical splitter 122 of the decision unit U n,i , and the control signal C n which is an optical pulse is sent to the decision unit U n,i . Deflect.
  • FIG. 5B shows another example of an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to this embodiment.
  • an optical resonator circuit 124 is provided to lead the optical pulse from the optical waveguide 121 to the determination unit U.
  • the optical resonator circuit 124 for example, a high-speed modulation optical resonator such as an optical disk resonator, an optical ring resonator, or the like, which has a small element size and can operate with low energy, can be applied.
  • FIG. 6A shows an opto-electric hybrid logic circuit of the decision unit according to this embodiment.
  • the mixed opto-electrical logic circuit 123 an electrical signal with controlled duration is generated from the light pulses directed to the decision unit U n,i .
  • An inverting amplifier A1 and a non-inverting amplifier A2 are connected to the output of the photodiode PD, which is a light-receiving element to which the control signal Cn , which is an optical pulse, is incident.
  • the output of transistor Tr1, driven by signal Ln -1,j from the previous decision stage S, is coupled through buffers A3 and A4 to transistors Tr2 and Tr3, respectively.
  • the non-inverting amplifier A2 drives the transistor Tr3 to output the signal Ln,2i-1 to one of the two output ports of the decision unit U. to output
  • the inverting amplifier A1 drives the transistor Tr2 to output the signal L n,2i to the other output port of the decision unit U.
  • FIG. 6B shows a truth table representing the input/output relationship of the opto-electric hybrid logic circuit.
  • the opto-electric hybrid logic circuit 123 includes a light receiving element and a determination circuit configured by an electronic circuit.
  • the decision circuit outputs the signal output from the preceding decision stage to one of the two output ports according to the truth table according to the output of the light receiving element.
  • a very fast pattern recognition circuit produces a spatially differentiated output signal in response to a combination of N input bits.
  • Very low-latency ultra-fast bit recognition is the basis for realizing the newly proposed processor, which uses this ultra-fast pattern recognition circuit to identify the operator that produces a particular output signal and the content of its operands. used to A calculation result corresponding to a combination of an operator and an operand can be obtained without executing calculation processing.
  • the processor 200 in the computing platform of this embodiment comprises a recognition circuit 202 that recognizes execution instructions including operators and operands output from the controller 201 .
  • Recognition circuit 202 includes, as described above, a serial-to-parallel converter that outputs a light pulse that controls the decision stage corresponding to each bit of the execution instruction, and the output of the decision stage that is selected by the light pulse to provide an execution instruction. and a decision circuit for outputting a control signal from a uniquely corresponding output port.
  • the recognition circuit 202 outputs a control signal uniquely corresponding to the operator and the operand to the memory 203 as a recognition result.
  • the memory 203 preliminarily stores calculation results according to the control signal. A search result corresponding to the control signal is returned to the controller 201 via the selection circuit 204 .
  • the processor 200 searches the memory 203 and obtains the already stored calculation result without performing actual calculation.
  • a computing platform capable of high-speed operation can be configured by applying the above-described ultra-high-speed pattern recognition circuit for recognition of execution instructions and combining it with a high-speed optical memory to be described below. Also, according to this configuration, the complexity of the execution instruction does not affect the speed at which the final calculation result is received, thus facilitating speeding up of calculation processing.
  • FIG. 8 shows the configuration of the computing platform of this embodiment.
  • a set of N parallel signals (N bits) are output from the controller 201 representing the instructions to be executed (operators + operands).
  • Execution instructions are fetched from a cache memory internal to controller 201, similar to the central processing unit (CPU) of a conventional von Neumann computing architecture.
  • the execution command is converted into a serial signal by a parallel-serial conversion circuit (P/S) 251 and input to the recognition circuit 202 .
  • P/S parallel-serial conversion circuit
  • the recognition circuit 202 is the ultra-high-speed pattern recognition circuit described above, and the execution command is input to the serial-parallel converter 110, and N light pulses are generated corresponding to the input N-bit input word. .
  • the optical pulses generated by each conversion channel 111 are input to the decision circuit 120 in each decision stage S via an optical waveguide 121 .
  • the duration of the electrical signal generated by the controller 201 is approximately N times the duration of the light pulse, as described above. For example, if each optical pulse represents 10 bits with a duration of 40 ps, the controller 201 can clock at 2.5 Gb/s and operate at the speed of common electronic circuits. By increasing the number of bits to which the light pulses correspond, the clock speed of the controller 201 can be further relaxed, resulting in higher computational throughput at the same clock speed.
  • the electrical signal output from the recognition circuit 202 drives the optical resonator 232 of the memory 203 to retrieve the calculation results pre-stored in the memory 203 . Details of the memory 203 will be described later.
  • a high-speed optical memory will be described as an example of the memory 203, but in the computing platform of this embodiment, a memory made up of ordinary electronic circuits can also be used.
  • the search result of memory 203 is output as a series of serial light pulses representing the calculation result.
  • the serial light pulses are routed by the optical switch 241 of the selection circuit 204 to the recognition circuit 202 to be sent back to the controller 201 or used as new operands in the next computational step.
  • Optical pulses sent back to the controller 201 are converted into electrical signals and output via a serial-to-parallel converter (S/P) 242 .
  • S/P serial-to-parallel converter
  • FIG. 9 shows a schematic configuration of the optical memory of the computing platform of this embodiment.
  • the memory 203 is an all-optical memory that processes everything from input to output with optical signals.
  • the memory 203 has 2 N optical resonators 232 corresponding to the 2 N output ports of the recognition circuit 202 (a case of four output ports is shown for convenience of explanation).
  • a plurality of optical resonators 232 a - d are optically coupled with an optical waveguide 237 connected to the optical pulse source 231 .
  • the driven optical resonator divides the optical pulse from the optical pulse light source 231. output to circuits 233a-d.
  • the splitting circuit 233 has optical paths for splitting an optical pulse into a large number of optical pulses, and each optical path is provided with a delay circuit 236 having a predetermined delay time.
  • the split light pulses are reflected or absorbed by LCOS (liquid crystal on silicon) 234, which is a spatial light phase modulator.
  • the reflected light pulse propagates along the original light path and is output as a series of serial light pulses through circulator 235 .
  • a series of serial light pulses representing a desired calculation result can be generated from one input light pulse (FIG. 9). can derive a light pulse representing the calculation result of "10110101". Since the execution instruction recognized by the recognition circuit 202 is optically processed in the memory 203, the response time can be shortened and the power consumption can be reduced as compared with a memory made up of a conventional electronic circuit.
  • FIG. 10 shows another configuration example of the optical memory of this embodiment.
  • An example of increasing the number of processing bits of a processor to obtain higher computational throughput is shown.
  • the number of output ports of the recognition circuit increases exponentially, so the configuration of the division circuit and LCOS in the optical memory also increases dramatically. Therefore, the memory 303 processes by wavelength division multiplexing using optical pulses of a plurality of wavelengths.
  • the optical pulse light source 331 is a multi-wavelength light source that emits optical pulses of multiple wavelengths. For example, an input word is given an additional bit for wavelength selection in addition to an execution command.
  • the recognition circuit 202 outputs an electric signal OUT as a final result and an additional bit WSEL for wavelength selection from one of 2 N output ports as a result of recognizing an execution command.
  • the optical pulse light source 331 selects one wavelength corresponding to the additional bit WSEL and outputs an optical pulse, and the optical resonator 332 to which the electrical signal OUT is input outputs the optical pulse of the selected wavelength to the division circuit. do.
  • An arrayed waveguide grating (AWG) 337 is inserted as a branching circuit in the preceding stage of the dividing circuit, and the optical pulse output from the optical resonator 332 is branched to the dividing circuits 333a-333d according to the wavelength.
  • the split light pulses are controlled in reflection by LCOS 334 and output as a series of serial light pulses through circulator 335 .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Optical Communication System (AREA)

Abstract

Provided is a computing platform wherein high speed operations can be achieved. This photonics/electronics integrated computer comprises a recognition circuit that recognizes an executive instruction including operators and operands outputted from a controller. The recognition circuit includes: a serial-to-parallel converter that outputs optical pulses for controlling the decision stages corresponding to the respective bits of the executive instruction; and a decision circuit that uses the optical pulses to select outputs of the decision stages and that outputs control signals from the output ports uniquely corresponding to the executive instruction. This photonics/electronics integrated computer further comprises a memory that outputs, in accordance with the control signals, calculation results stored therein beforehand.

Description

光電子融合型コンピュータoptoelectronic computer
 本発明は、光電子融合型コンピュータに関し、より詳細には、光パルスと電気パルスとを組み合わせたコンピューティングプラットフォームを備えた光電子融合型コンピュータに関する。 The present invention relates to an optoelectronic convergence computer, and more particularly to an optoelectronic convergence computer provided with a computing platform that combines optical pulses and electrical pulses.
 デジタル信号処理のためのチップは、非常に多くの論理ゲートが高密度に集積された電子回路を含む。伝統的に、これらの電子回路は、相補型金属酸化膜半導体(CMOS)トランジスタで実現されている。CMOSトランジスタの微細化が進み、集積密度の高い電子回路が実現され、クロック速度の高速化が進められ、信号処理の機能、性能が継続的に向上している。しかしながら、現在のCMOSトランジスタは、数ナノメートルの寸法まで微細化が進み、さらなる小型化は、様々な技術的課題および高コスト化のために困難になっている。 A chip for digital signal processing contains electronic circuits densely integrated with a large number of logic gates. Traditionally, these electronic circuits have been implemented with complementary metal oxide semiconductor (CMOS) transistors. The miniaturization of CMOS transistors has progressed, electronic circuits with high integration densities have been realized, the clock speed has been increased, and the function and performance of signal processing have been continuously improved. However, current CMOS transistors have been miniaturized to dimensions of several nanometers, and further miniaturization has become difficult due to various technical issues and high costs.
 この問題を解決するために、例えば、神経形態学的スキーム、量子計算スキームなどの代替的なコンピューティングアーキテクチャが注目されている。このようなパラダイムシフトは、特定の計算問題に対しては、解決策を提供することに成功している。一方で、従来のフォン・ノイマン型コンピューティングアーキテクチャの性能を向上させる要求も存在する。 To solve this problem, attention is focused on alternative computing architectures, such as neuromorphic schemes and quantum computation schemes. Such paradigm shifts have successfully provided solutions to certain computational problems. On the other hand, there is also a demand to improve the performance of traditional von Neumann computing architectures.
特許6701443号公報Japanese Patent No. 6701443
 このような要求を満たすために、本発明の目的は、光パルスと電気パルスとを組み合わせて、高速動作が可能なコンピューティングプラットフォームを備えた光電子融合型コンピュータを提供することにある。 In order to meet such demands, the object of the present invention is to provide an opto-electronic convergence computer equipped with a computing platform capable of high-speed operation by combining optical pulses and electrical pulses.
 本発明は、このような目的を達成するために、光電子融合型コンピュータの一実施態様は、コントローラから出力されるオペレータとオペランドとを含む実行命令を認識する認識回路であって、前記実行命令のビットごとに対応する決定段階を制御する光パルスを出力するシリアル-パラレル変換器、および前記光パルスにより前記決定段階の出力が選択されて、前記実行命令に一意に対応した出力ポートから制御信号を出力する決定回路を含む認識回路と、前記制御信号に応じて予め記憶されている計算結果を出力するメモリとを備えたことを特徴とする。 In order to achieve such an object, the present invention provides an embodiment of an opto-electronic integrated computer, which is a recognition circuit for recognizing an execution instruction including an operator and an operand output from a controller, comprising: a serial-to-parallel converter outputting a light pulse that controls a bit-by-bit corresponding decision stage, and said light pulse selects the output of said decision stage to drive a control signal from an output port uniquely corresponding to said execution instruction. It is characterized by comprising a recognition circuit including a decision circuit for outputting, and a memory for outputting a pre-stored calculation result according to the control signal.
図1は、本発明の一実施形態にかかる超高速パターン認識回路における信号処理を示す図、FIG. 1 is a diagram showing signal processing in an ultra-high-speed pattern recognition circuit according to one embodiment of the present invention; 図2は、本実施形態にかかる4ビットのワードを処理可能な超高速パターン認識回路の構成を示す図、FIG. 2 is a diagram showing the configuration of an ultra-high-speed pattern recognition circuit capable of processing 4-bit words according to this embodiment; 図3は、4ビットで表される入力ワードを10進数で表される出力に対応付ける方法を説明する図、FIG. 3 is a diagram explaining how to correspond an input word represented by 4 bits to an output represented by a decimal number; 図4は、本実施形態にかかる超高速パターン認識回路における処理シーケンスを示す図、FIG. 4 is a diagram showing a processing sequence in the ultra-high-speed pattern recognition circuit according to this embodiment; 図5Aは、本実施形態にかかる超高速パターン認識回路の決定ユニットとして用いる光電気混載回路の一例を示す図、FIG. 5A is a diagram showing an example of an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to the present embodiment; 図5Bは、本実施形態にかかる超高速パターン認識回路の決定ユニットとして用いる光電気混載回路の他の例を示す図、FIG. 5B is a diagram showing another example of an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to the present embodiment; 図6Aは、本実施形態にかかる決定ユニットの光電気混載論理回路を示す図、FIG. 6A is a diagram showing an opto-electric hybrid logic circuit of a decision unit according to the present embodiment; 図6Bは、光電気混載論理回路の入出力の関係を表す真理値表を示す図、FIG. 6B is a diagram showing a truth table representing the input/output relationship of the opto-electric hybrid logic circuit; 図7は、本発明の一実施形態にかかるコンピューティングプラットフォームの原理を説明するための図、FIG. 7 is a diagram for explaining the principle of a computing platform according to one embodiment of the present invention; 図8は、本実施形態のコンピューティングプラットフォームの構成を示す図、FIG. 8 is a diagram showing the configuration of the computing platform of this embodiment; 図9は、本実施形態のコンピューティングプラットフォームの光メモリの概略の構成を示す図、FIG. 9 is a diagram showing a schematic configuration of the optical memory of the computing platform of this embodiment; 図10は、本実施形態の光メモリの他の構成例を示す図である。FIG. 10 is a diagram showing another configuration example of the optical memory of this embodiment.
 以下、図面を参照しながら本発明の実施形態について詳細に説明する。光パルスは、低ジッタで生成することができ、低損失および低分散で長距離を伝搬することができる。また、光パルスが伝播する光導波路の長さを調整することにより、光パルスのタイミングを正確に制御することができる。例えば、20μmの光ファイバは、約0.1psの遅延を生じる。このような特性により、高速のデジタル信号処理に適しているが、CMOSトランジスタからなる論理ゲートとは異なり、光論理ゲートを大規模に集積することは、困難である。そこで、光パルスと電気パルスとを組み合わせた光電子融合型のコンピューティングプラットフォームを構築する。 Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Optical pulses can be generated with low jitter and can propagate long distances with low loss and dispersion. Also, by adjusting the length of the optical waveguide through which the optical pulse propagates, the timing of the optical pulse can be controlled accurately. For example, a 20 μm optical fiber produces a delay of approximately 0.1 ps. These characteristics make it suitable for high-speed digital signal processing, but unlike logic gates made up of CMOS transistors, it is difficult to integrate optical logic gates on a large scale. Therefore, we will construct an optoelectronic convergence computing platform that combines optical and electrical pulses.
  [超高速パターン認識回路]
 図1に、本発明の一実施形態にかかる超高速パターン認識回路における信号処理を示す。認識回路100は、高速のビット信号が入力される1つのシリアルポートと、Nビットの入力ワードを認識し、2N通りのビット組み合わせに一意対応する2N個の出力ポートとを備えている。認識回路100の出力ポートは、電気回路130の2N個の入力ポートに接続されている(例えば、特許文献1参照)。
[Ultra-high-speed pattern recognition circuit]
FIG. 1 shows signal processing in an ultra-high speed pattern recognition circuit according to one embodiment of the present invention. The recognition circuit 100 has one serial port to which a high-speed bit signal is input, and 2 N output ports that recognize an N-bit input word and uniquely correspond to 2 N bit combinations. The output ports of the recognition circuit 100 are connected to 2 N input ports of the electrical circuit 130 (see, for example, Patent Document 1).
 シリアルポートに信号が入力されていない状態では全ての出力は常にLowレベルの状態にある。ワードを構成するビット信号が認識回路100のシリアルポートに入力されると、そのワードのビット組み合わせに対応する1つの出力ポートだけがHighレベルの状態になる。認識回路100はリアルタイムで作動し、Nビットのワードが入力されると直ちに対応する出力ポートがHighレベルになり、出力を低速な電気回路に接続するのに十分な持続時間の間、Highレベルの状態を保つ。 All outputs are always at the low level when no signal is input to the serial port. When a bit signal constituting a word is input to the serial port of the recognition circuit 100, only one output port corresponding to the bit combination of the word becomes High level. Recognition circuit 100 operates in real-time such that as soon as an N-bit word is input, the corresponding output port goes high and stays high for a duration sufficient to connect the output to a slow electrical circuit. keep state.
 認識回路100は、入力されたワードのビット組み合わせに基づいて、ビット組み合わせに対応する空間的に分離された出力ポートに出力信号を生成する。この出力ポートから出力される信号の持続時間は、1つのワード分の持続時間(Tword)、すなわち全てのビット信号分の持続時間に対応しているため、低速な電気回路の速度に合わせることが十分可能な長さである。この場合、生成した信号が出力される出力ポートには全てのビット信号の集合情報、すなわちビット組み合わせの情報が含まれるため、従来と較べてクロック周期を消費せずに済み、電気回路での処理を簡素化することができる。 Recognition circuit 100 generates output signals at spatially separated output ports corresponding to the bit combinations based on the bit combinations of the input word. The duration of the signal output from this output port corresponds to the duration of one word (T word ), i.e. the duration of all the bit signals, so it can be adapted to the speed of slow electrical circuits. is sufficiently long. In this case, since the output port to which the generated signal is output contains the collective information of all bit signals, that is, the information of the bit combination, it is possible to consume less clock cycles than in the past, and the processing in the electric circuit is reduced. can be simplified.
 複数の連続したNビットのワードが、論理演算を行う必要のあるM個のワードとなって任意の時間間隔(Tnext)で入力されると、各ワードをビット組み合わせに応じて空間的に異なる出力ポートに時間間隔(Tnext)で出力される。このM個の連続した低速な信号を処理することのできる論理回路が演算を実行する場合を考える。M個の各ワードに対応する出力は、連続的に電気回路130に入力される。最初の入力が最初の処理結果となるが、次の入力はその1つ前の結果を用いて処理され、全ての演算が終了するまで更新される。これにより、低速な電気回路で超高速なビットの処理時間を削減することが可能な、ワード毎の処理方式を確立する。 When a number of consecutive N-bit words are input at arbitrary time intervals (T next ) into M words that need to be logically operated on, each word is spatially distinct according to its bit combination. It is output to the output port at time intervals (T next ). Consider the case where a logic circuit capable of processing these M consecutive slow signals performs the operation. Outputs corresponding to each of the M words are sequentially input to electrical circuit 130 . The first input is the first processed result, but the next input is processed using the previous result and updated until all operations are completed. This establishes a word-by-word processing method that can reduce the processing time of ultra-high-speed bits with a low-speed electric circuit.
 図2に、本実施形態にかかる4ビットのワードを処理可能な超高速パターン認識回路の構成を示す。認識回路100は、シリアル-パラレル変換器110および決定回路120の2つの主要な機能ブロックで構成される。シリアル-パラレル変換器110の各変換チャネルの出力は、決定回路120の所定の決定段階Sを制御する。 FIG. 2 shows the configuration of an ultra-high-speed pattern recognition circuit capable of processing 4-bit words according to this embodiment. Recognition circuit 100 consists of two major functional blocks, serial-to-parallel converter 110 and decision circuit 120 . The output of each conversion channel of serial-to-parallel converter 110 controls a predetermined decision stage S of decision circuit 120 .
 最も重要な最上位ビットは、その状態がHighレベルかLowレベルかにより、最終的な出力が8未満か、8以上かを決定する。従って、最上位のビットの値が分かれば、最終出力が取り得る値の候補を半分にすることができる。残った値の候補についても、次に上位のビットの状態が分かればそれを半分にすることができ、つまり最終出力の値の候補を1/4に絞り込むことができる。この手順を最下位ビットまで繰り返すことにより、適切でない出力の可能性を連続的に排斥し、ワードを正しい出力に変換する、すなわちワードのビット組み合わせに対応した出力ポートからの出力のみをHighレベルの状態にすることが可能となる。このことから、決定回路120は下記のように構成する。 The most significant bit determines whether the final output is less than 8 or greater than 8 depending on whether the state is High level or Low level. Therefore, if the value of the most significant bit is known, it is possible to halve the possible values of the final output. If the state of the next higher-order bit is known, the remaining value candidates can also be halved, that is, the final output value candidates can be narrowed down to 1/4. By repeating this procedure down to the least significant bit, the possibility of improper output is continuously eliminated, and the word is converted to the correct output, that is, only the output from the output port corresponding to the bit combination of the word is set to a high level. state can be made. Therefore, the decision circuit 120 is configured as follows.
 最上位ビットに対応する決定段階S1は決定ユニットU1,1を1つ含み、最上位ビットから1つ下位のビットに対応する決定段階S2は2つの決定ユニットU2,1、U2,2を含み、さらに1つ下位のビットに対応する決定段階S3は4つの決定ユニットU3,1~U3,4を含む。最下位ビットに対応する決定段階S4は、8つの決定ユニットU4,1~U4,8を含む。 The decision stage S1 corresponding to the most significant bit includes one decision unit U 1,1 , and the decision stage S2 corresponding to the bit below the most significant bit comprises two decision units U 2,1 , U 2,2 . , and the decision stage S3 corresponding to the one lower bit includes four decision units U 3,1 to U 3,4 . The decision stage S4 corresponding to the least significant bit comprises eight decision units U 4,1 to U 4,8 .
 最上位のビットに対応する決定段階S1における決定ユニットU1,1の2つの出力ポートは、一方が2番目に上位のビットに対応する決定段階S2の決定ユニットU2,1に接続され、他方が決定ユニットU2,2に接続されている。同様に、決定段階S2の決定ユニットU2,1、U2,2の4つの出力ポートは、3番目に上位のビットに対応する決定段階S3の4つの決定ユニットU3,1~U3,4に接続され、決定段階S3の4つの決定ユニットU3,1~U3,4の8つの出力ポートは、最下位のビットに対応する決定段階S4の8つの決定ユニットU4,1~U4,8に接続されている。 The two output ports of the decision unit U 1,1 in the decision stage S1 corresponding to the most significant bit are connected to the decision unit U 2,1 of the decision stage S2 corresponding to the second most significant bit, the other is connected to the decision unit U 2,2 . Similarly, the four output ports of decision units U 2,1 , U 2,2 of decision stage S2 are connected to the four decision units U 3,1 to U 3, of decision stage S3 corresponding to the third most significant bit. 4 , the eight output ports of the four decision units U 3,1 to U 3,4 of the decision stage S3 are connected to the eight decision units U 4,1 to U of the decision stage S4 corresponding to the least significant bits. Connected to 4,8 .
 最初の決定段階S1の決定ユニットU1,1は、シリアル-パラレル変換器110の最上位のビット信号を変換する変換チャネルで生成された制御信号C1に基づいて、2つある出力ポートのうちの一方をHighレベルにする。決定ユニットU1,1の出力の一方がHighレベルにされると、2段目の決定段階S2の2つの決定ユニットU2,1、U2,2のうちの一方だけをアクティブ化する。そのアクティブ化された決定ユニットU2,1又はU2,2は、シリアル-パラレル変換器110の2番目に上位のビット信号を変換する変換チャネルで生成された制御信号C2に基づいて、2つある出力ポートのうちの一方をHighレベルにする。これら処理により、2段目の決定段階S2が選び得る4つの出力ポートのうちの1つだけがHighレベルになり、さらに空間的に分離された出力ポートに信号が生成されて、最終出力が取り得る可能性を1/4に絞り込む。同様に、2段目の決定段階S2からのHighレベルの信号によって、3段目の決定段階S3の4つの決定ユニットU3,1~U3,4のうちの1つがアクティブ化される。そしてそのアクティブ化された決定ユニットU3,1~U3,4のうちの1つは、シリアル-パラレル変換器110の3番目に上位のビット信号を変換する変換チャネルで生成された制御信号C3に基づいて、2つある出力ポートのうちの一方をHighレベルにする。4段目の決定段階S4の8つの決定ユニットU4,1~U4,8もいずれか1つが3段目の決定段階S3からのHighレベルの信号によってアクティブ化され、シリアル-パラレル変換器110の最下位のビット信号を変換する変換チャネルで生成された制御信号C4に基づいて、2つある出力ポートのうちの一方をHighレベルにする。 The decision unit U 1,1 of the first decision stage S 1 selects one of the two output ports based on the control signal C 1 generated by the conversion channel that converts the most significant bit signal of the serial-parallel converter 110 . One is set to High level. When one of the outputs of the decision unit U 1,1 is brought to a high level, it activates only one of the two decision units U 2,1 , U 2,2 of the second decision stage S2. The activated decision unit U 2,1 or U 2,2 makes two decisions based on the control signal C 2 generated in the conversion channel converting the second most significant bit signal of the serial-to-parallel converter 110 . One of certain output ports is set to High level. Through these processes, only one of the four output ports that can be selected by the second decision step S2 becomes High level, and signals are generated at spatially separated output ports to obtain the final output. Narrow your chances to 1/4. Similarly, a high level signal from the second decision stage S2 activates one of the four decision units U 3,1 to U 3,4 of the third decision stage S3. One of the activated decision units U 3,1 to U 3,4 then receives the control signal C 3 generated in the conversion channel converting the third most significant bit signal of the serial-to-parallel converter 110 . , one of the two output ports is set to High level. Any one of the eight decision units U 4,1 to U 4,8 in the fourth decision stage S4 is activated by the High level signal from the third decision stage S3, and the serial-parallel converter 110 One of the two output ports is set to High level based on the control signal C4 generated by the conversion channel that converts the least significant bit signal of .
 図3に示すように、4ビットで表される入力ワードを10進数で表される出力に対応付け、16個の出力ポートのうちの1つの出力をHighレベルにすることができる。例えば、4ビットのワード「1101」を認識回路100に入力した場合、異なるバイナリの組み合わせを、十進数の0から15の整数の1つである「11」に対応させることができる。 As shown in FIG. 3, an input word represented by 4 bits can be associated with an output represented by a decimal number, and the output of one of the 16 output ports can be set to a high level. For example, if the 4-bit word "1101" is input to the recognition circuit 100, different binary combinations may correspond to "11", one of the decimal integers 0-15.
 認識回路100は、上記4ビットのワードの処理に限定されず、任意のビット組み合わせを有するNビットのワードについても、上記手順を同様に繰り返し実行することで、ワード毎のビット組み合わせに対応する出力ポートの出力のみをHighレベルの状態することができる。このとき、シリアル-パラレル変換器110はワードを構成する各ビットに対応するN個の変換チャネルを有し、決定回路120はシリアル-パラレル変換器110のN個の変換チャネルに対応するN段の決定段階S1~SNを有する構成とする。 The recognition circuit 100 is not limited to the processing of the 4-bit words described above. By repeatedly executing the above procedure for N-bit words having arbitrary bit combinations, the output corresponding to the bit combinations for each word is obtained. Only the output of the port can be brought to a high level state. At this time, the serial-parallel converter 110 has N conversion channels corresponding to each bit constituting a word, and the decision circuit 120 has N stages corresponding to the N conversion channels of the serial-parallel converter 110. It is assumed that there are decision stages S 1 to S N .
 N段の決定段階S1~SNは、最上位のビットに対応する決定段階S1を1段目とすると、2s(s=0、1、2、・・・、N-1)の位のビットに対応する(N-i)段目の決定段階S(N-s)は、2N-1-s個の決定ユニットUを含み、N番目の決定段階SNは2N-1個の決定ユニットUN-s,1~UN-s,t(t=2N-1)で構成される。 The N stages of decision steps S 1 to S N are 2 s (s=0, 1 , 2, . The (N-i)th decision stage S (Ns) corresponding to the bit of the order includes 2 N-1-s decision units U, and the N-th decision stage S N includes 2 N-1 It is composed of decision units U Ns,1 to U Ns,t (t=2 N-1 ).
 各決定ユニットUは、2つの出力ポートを有しており、各出力ポートは、1つ下位の決定段階Sにある異なる決定ユニットUに1対1でそれぞれ接続し、そのうちの一方の決定ユニットUだけをアクティブ化する。各決定段階Sでは、1つ上位の決定段階SからのHighレベルの出力によって一度に1つの決定ユニットUだけがアクティブ化される。 Each decision unit U has two output ports, each output port is connected one-to-one to a different decision unit U in the next lower decision stage S, one of which U only activate. At each decision stage S, only one decision unit U is activated at a time by a high level output from the decision stage S above it.
 同じ決定段階Sに属する決定ユニットUは、シリアル-パラレル変換器110の同一の変換チャネルに並列に接続されており、アクティブ化された決定ユニットUの出力は、その決定ユニットUが属する決定段階Sに対応するシリアル-パラレル変換器の変換チャネルで生成された制御信号Cで制御される。制御信号Cの状態がHighレベルのとき、アクティブ化された決定ユニットUの一方のポートだけがHighレベルになり、変換されたビット信号がlowレベルのとき、今度は他方のポートだけがHighレベルになる。 The decision units U belonging to the same decision stage S are connected in parallel to the same conversion channel of the serial-to-parallel converter 110, and the output of the activated decision unit U is the decision stage S to which the decision unit U belongs. is controlled by a control signal C generated in the conversion channel of the serial-to-parallel converter corresponding to . When the state of the control signal C is high level, only one port of the activated decision unit U is high level, and when the converted bit signal is low level, only the other port is high level in turn. Become.
 このようにして各決定段階Sに属する決定ユニットUのうちの1つの決定ユニットUの一方の出力のみがHighレベルになり、認識回路100の最終出力として、N番目の決定段階SNの決定ユニットUN-s,1~UN-s,t(t=2N-1)の2N個の出力ポートのうち、Nビットのワードのビット組み合わせに対応する1つをHighレベルにすることができる。 In this way, only one output of one of the decision units U belonging to each decision stage S becomes High level, and the final output of the recognition circuit 100 is the decision unit of the N-th decision stage S N . One of 2 N output ports U Ns,1 to U Ns,t (t=2 N-1 ) corresponding to a bit combination of an N-bit word can be set to a high level.
 図4に、本実施形態にかかる超高速パターン認識回路における処理シーケンスを示す。制御信号C1はユニットU1,1の出力だけを決定し、本例では最上位ビットがHighレベルであることから、信号L1,1をHighレベルに変換する。一方で、信号L1,2は定常状態のLowレベルのままである。本回路は、繰返し演算することを考慮して信号L1,1の持続時間を4T(T:クロックサイクル時間)に設定しており、新しいワードが時間4Tより後に到着すると、ユニットU1,1の出力を再び自由に決定することができる。各ユニットの出力にとって重要な機能は、十分に早い立ち上がり時間であり、これは回路全体の演算を行うのに不可欠なものである。 FIG. 4 shows a processing sequence in the ultra-high speed pattern recognition circuit according to this embodiment. Control signal C1 determines only the output of unit U 1,1 and converts signal L 1,1 to a high level since the most significant bit is a high level in this example. On the other hand, the signal L 1,2 remains at a steady-state low level. The circuit sets the duration of signal L 1,1 to 4T (T: clock cycle time) to allow for repeated operations, and when a new word arrives after time 4T, unit U 1,1 can again be freely determined. An important feature for each unit's output is a sufficiently fast rise time, which is essential for the operation of the entire circuit.
 信号L1,1がHighレベルになると、それに応じて信号L2,2もHighレベルになる。しかし、制御信号C2がHighレベルで生成されると、信号L2,2はリセットされ、代わりに信号L2,1がHighレベルになる。ここで、3番目に上位のビットがLowだと、制御信号C3はLowレベルになり、信号L3,2は信号L2,1で初期化された後はHighレベルのままになる。信号L4,3は制御信号C4が生成された後はHighレベルになり、回路の最終出力を生成する。 When the signal L 1,1 becomes High level, the signal L 2,2 also becomes High level accordingly. However, when control signal C2 is generated at a high level, signal L2,2 is reset and signal L2,1 goes high instead. Here, when the third most significant bit is Low, the control signal C3 becomes Low level, and the signal L3,2 remains High level after being initialized by the signal L2,1 . Signal L4,3 goes high after control signal C4 is generated, producing the final output of the circuit.
 所定の段階SnにおいてクロックパルスKnの時間を基準とすると、この段階の出力開始時間にばらつきが生じる。例では、制御信号C3が発生する前に信号L3,2が始まるが、しかし、3番目に上位のビット信号がhighレベルとなる場合(本例では変換されたビット信号はlowレベル)、制御信号L3,1はhighレベルとなり信号C3の開始よりもわずか後に始まる。このように、各段階において出力開始のばらつきがあると、回路の最終出力において信号の持続時間に影響を及ぼす。 If the time of the clock pulse Kn is used as a reference at a predetermined stage Sn, variations occur in the output start time of this stage. In the example, signal L 3,2 begins before control signal C3 occurs, but if the third most significant bit signal goes high (in this example the converted bit signal is low), control Signal L 3,1 goes high starting slightly after the start of signal C3. Thus, any variation in output onset at each stage affects the duration of the signal at the final output of the circuit.
 各決定段階Sにおいて発生した信号は、次の決定段階にある決定ユニットUのうち、1つの決定ユニットUのみを制御するために使用される。すなわち、信号はごく僅かな数のトランジスタを、速い動作を妨げる電気負荷の発生を伴わずに動かす必要がある。また、このような構成において高速電気信号の処理を可能とするために、処理中の信号速度に対応できるような寸法で集積回路を設計しなければならない。 A signal generated in each decision stage S is used to control only one decision unit U among the decision units U in the next decision stage. That is, the signal needs to move very few transistors without creating an electrical load that prevents fast operation. Also, to allow processing of high speed electrical signals in such an arrangement, the integrated circuit must be designed with dimensions to accommodate the speed of the signals being processed.
 一方、特定の決定段階Sへシリアル-パラレル変換器110が発する各制御信号Cは、その決定段階に属する全ての決定ユニットを制御しなければならない。従来の電子回路においては、制御信号を大規模な容量性負荷を伴う数多くのトランジスタへ繋げる必要があり、素早い動作を妨げている。また、容量性負荷が大きい場合は、必要以上に立ち上がり時間が長くなり、信号の持続時間を伸ばしてしまう。そこで、以下で説明する光パルスと電気パルスとを組み合わせた光電気混載回路を適用する。 On the other hand, each control signal C n issued by the serial-to-parallel converter 110 to a particular decision stage S n must control all decision units belonging to that decision stage. In conventional electronic circuits, the control signals must be routed through numerous transistors with large capacitive loads, hindering rapid operation. Also, if the capacitive load is large, the rise time will be longer than necessary, extending the duration of the signal. Therefore, an opto-electric hybrid circuit combining an optical pulse and an electric pulse, which will be described below, is applied.
 図5Aに、本実施形態にかかる超高速パターン認識回路の決定ユニットとして用いる光電気混載回路を示す。シリアル-パラレル変換器110の各変換チャネル111の出力を、決定段階Sに含まれる決定ユニット群の列に近接して配置される光導波路121に接続する。決定ユニットUは、1つ前の決定段階Sから決定ユニットUをアクティブ化するための信号Ln-1,jに従って、光パルスを光導波路121から決定ユニットUに引き込むための光スプリッタ122と、電気パルスおよび光パルスの2つの入力信号に従って2つの出力ポートのどちらか一方から電気パルスを出力するかを決定する光電気混載論理回路123とから構成される。 FIG. 5A shows an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to this embodiment. The output of each conversion channel 111 of the serial-to-parallel converter 110 is connected to an optical waveguide 121 located adjacent to the row of decision units involved in the decision stage S. the decision unit U includes an optical splitter 122 for drawing light pulses from the optical waveguide 121 into the decision unit U according to the signal L n-1,j for activating the decision unit U from the previous decision stage S; and an opto-electric hybrid logic circuit 123 which determines whether to output an electric pulse from one of the two output ports according to two input signals of an electric pulse and an optical pulse.
 変換されたビット信号がhighである場合は、シリアル-パラレル変換器110の変換チャネル111から決定段階Sの決定ユニットUに向けて光パルスである制御信号Cnが発生され、一方、変換されたビット信号がlowである場合、光パルスは発生しない。 If the converted bit signal is high, a control signal C n which is a light pulse is generated from the conversion channel 111 of the serial-to-parallel converter 110 towards the decision unit U of the decision stage S, while the converted If the bit signal is low, no light pulse is generated.
 信号Ln-1,jは、決定ユニットUn,iをアクティブ化するために使用されている。ここでは、決定ユニットUn,iを含む決定段階Sの全決定ユニットUに対して個別に対応する制御信号Cnを出力するのではなく、光パルスである制御信号Cnをアクティブ化された決定ユニットUn,iへ偏向させるために信号Ln-1,iを用いる。信号Ln-1,iを分岐して、信号Ln-1,iで決定ユニットUn,iの光スプリッタ122を制御し、光パルスである制御信号Cnを決定ユニットUn,iへ偏向させる。 The signal L n-1,j is used to activate the decision unit U n,i . Here, instead of outputting individually corresponding control signals C n for all decision units U n of the decision stage Sn containing the decision unit U n ,i , the control signals C n which are light pulses are activated. The signal L n-1,i is used to deflect to the determined decision unit U n,i . The signal L n-1,i is split, the signal L n-1,i controls the optical splitter 122 of the decision unit U n,i , and the control signal C n which is an optical pulse is sent to the decision unit U n,i . Deflect.
 図5Bに、本実施形態にかかる超高速パターン認識回路の決定ユニットとして用いる光電気混載回路の他の例を示す。光パルスを光導波路121から決定ユニットUに引き込むために、光スプリッタ122に代えて、光共振器回路124を備えている。光共振器回路124としては、例えば、光ディスク共振器、光リング共振器など、素子のサイズが小さく、低エネルギーで動作が可能な高速変調光共振器を適用することができる。 FIG. 5B shows another example of an opto-electric hybrid circuit used as a decision unit of the ultra-high-speed pattern recognition circuit according to this embodiment. In place of the optical splitter 122, an optical resonator circuit 124 is provided to lead the optical pulse from the optical waveguide 121 to the determination unit U. As the optical resonator circuit 124, for example, a high-speed modulation optical resonator such as an optical disk resonator, an optical ring resonator, or the like, which has a small element size and can operate with low energy, can be applied.
 図6Aに、本実施形態にかかる決定ユニットの光電気混載論理回路を示す。光電気混載論理回路123では、決定ユニットUn,iに偏向された光パルスから、制御された持続性を持つ電気信号を発生させる。光パルスである制御信号Cnが入射される受光素子であるフォトダイオードPDの出力に、反転増幅器A1と非反転増幅器A2とが接続されている。1つ前の決定段階Sからの信号Ln-1,jによって駆動されるトランジスタTr1の出力は、バッファA3,A4を介して、それぞれトランジスタTr2,Tr3に接続されている。制御信号Cnとして光パルスがフォトダイオードPDに入射されると、非反転増幅器A2がトランジスタTr3を駆動して、決定ユニットUの2つある出力ポートのうちの一方に信号Ln,2i-1を出力する。光パルスがフォトダイオードPDに入射されていない場合は、反転増幅器A1がトランジスタTr2を駆動して、決定ユニットUの他方の出力ポートに信号Ln,2iを出力する。 FIG. 6A shows an opto-electric hybrid logic circuit of the decision unit according to this embodiment. In the mixed opto-electrical logic circuit 123 an electrical signal with controlled duration is generated from the light pulses directed to the decision unit U n,i . An inverting amplifier A1 and a non-inverting amplifier A2 are connected to the output of the photodiode PD, which is a light-receiving element to which the control signal Cn , which is an optical pulse, is incident. The output of transistor Tr1, driven by signal Ln -1,j from the previous decision stage S, is coupled through buffers A3 and A4 to transistors Tr2 and Tr3, respectively. When a light pulse is incident on the photodiode PD as the control signal Cn , the non-inverting amplifier A2 drives the transistor Tr3 to output the signal Ln,2i-1 to one of the two output ports of the decision unit U. to output When no light pulse is incident on the photodiode PD, the inverting amplifier A1 drives the transistor Tr2 to output the signal L n,2i to the other output port of the decision unit U.
 図6Bに、光電気混載論理回路の入出力の関係を表す真理値表を示す。光電気混載論理回路123は、受光素子と電子回路によって構成された判定回路とを含む。判定回路は、前段の決定段階から出力された信号を、受光素子の出力に応じて、2つの出力ポートのうちのいずれかに、真理値表に従った信号を出力する。 FIG. 6B shows a truth table representing the input/output relationship of the opto-electric hybrid logic circuit. The opto-electric hybrid logic circuit 123 includes a light receiving element and a determination circuit configured by an electronic circuit. The decision circuit outputs the signal output from the preceding decision stage to one of the two output ports according to the truth table according to the output of the light receiving element.
  [コンピューティングプラットフォーム]
 超高速パターン認識回路は、N個の入力ビットの組み合わせに応答して空間的に微分された出力信号を生成する。非常に低い待ち時間の超高速ビットの認識は、新たに提案するプロセッサを実現するための基礎であり、この超高速パターン認識回路を、特定の出力信号を生成するオペレータとそのオペランドの内容を識別するために使用する。オペレータとオペランドの組み合わせに対応する計算結果は、計算処理を実行することなく取得することができる。
[Computing Platform]
A very fast pattern recognition circuit produces a spatially differentiated output signal in response to a combination of N input bits. Very low-latency ultra-fast bit recognition is the basis for realizing the newly proposed processor, which uses this ultra-fast pattern recognition circuit to identify the operator that produces a particular output signal and the content of its operands. used to A calculation result corresponding to a combination of an operator and an operand can be obtained without executing calculation processing.
 図7を参照して、本発明の一実施形態にかかるコンピューティングプラットフォームの原理を説明する。本実施形態のコンピューティングプラットフォームにおけるプロセッサ200は、コントローラ201から出力されるオペレータとオペランドとを含む実行命令を認識する認識回路202を備える。認識回路202は、上述したように、実行命令のビットごとに対応する決定段階を制御する光パルスを出力するシリアル-パラレル変換器と、光パルスにより決定段階の出力が選択されて、実行命令に一意に対応した出力ポートから制御信号を出力する決定回路とを含む。認識回路202は、オペレータとオペランドとに一意に対応した制御信号を、認識結果としてメモリ203に出力する。メモリ203には、制御信号に応じた計算結果が予め記憶されている。制御信号に対応する検索結果は、選択回路204を介して、コントローラ201に返される。 The principle of a computing platform according to one embodiment of the present invention will be described with reference to FIG. The processor 200 in the computing platform of this embodiment comprises a recognition circuit 202 that recognizes execution instructions including operators and operands output from the controller 201 . Recognition circuit 202 includes, as described above, a serial-to-parallel converter that outputs a light pulse that controls the decision stage corresponding to each bit of the execution instruction, and the output of the decision stage that is selected by the light pulse to provide an execution instruction. and a decision circuit for outputting a control signal from a uniquely corresponding output port. The recognition circuit 202 outputs a control signal uniquely corresponding to the operator and the operand to the memory 203 as a recognition result. The memory 203 preliminarily stores calculation results according to the control signal. A search result corresponding to the control signal is returned to the controller 201 via the selection circuit 204 .
 プロセッサ200は、計算処理中の実行命令(オペレータ+オペランド)を、認識回路202によって認識すると、実際の計算は行わずに、メモリ203を検索して、既に記憶された計算結果を取得する。実行命令の認識には、上述した超高速パターン認識回路を適用し、以下に説明する高速の光メモリと組み合わせることにより、高速動作が可能なコンピューティングプラットフォームを構成することができる。また、この構成によれば、実行命令の複雑さは、最終的な計算結果を受け取る速度に影響を与えないので、計算処理の高速化が容易になる。 When the recognition circuit 202 recognizes an execution instruction (operator + operand) during calculation processing, the processor 200 searches the memory 203 and obtains the already stored calculation result without performing actual calculation. A computing platform capable of high-speed operation can be configured by applying the above-described ultra-high-speed pattern recognition circuit for recognition of execution instructions and combining it with a high-speed optical memory to be described below. Also, according to this configuration, the complexity of the execution instruction does not affect the speed at which the final calculation result is received, thus facilitating speeding up of calculation processing.
  [光電子融合型コンピュータ]
 図8に、本実施形態のコンピューティングプラットフォームの構成を示す。実行命令(オペレータ+オペランド)を表すN個のパラレル信号(Nビット)のセットが、コントローラ201から出力される。実行命令は、従来のフォン・ノイマン型コンピューティングアーキテクチャの中央処理装置(CPU)と同様に、コントローラ201内部のキャッシュメモリからフェッチされる。実行命令は、パラレル-シリアル変換回路(P/S)251により、シリアル信号に変換されて認識回路202に入力される。
[Optical-electronic convergence computer]
FIG. 8 shows the configuration of the computing platform of this embodiment. A set of N parallel signals (N bits) are output from the controller 201 representing the instructions to be executed (operators + operands). Execution instructions are fetched from a cache memory internal to controller 201, similar to the central processing unit (CPU) of a conventional von Neumann computing architecture. The execution command is converted into a serial signal by a parallel-serial conversion circuit (P/S) 251 and input to the recognition circuit 202 .
 認識回路202は、上述した超高速パターン認識回路であり、実行命令は、シリアル-パラレル変換器110に入力され、入力されたNビットの入力ワードに対応してN個の光パルスが生成される。各変換チャネル111で生成された光パルスは、光導波路121を介して、決定段階Sごとに決定回路120に入力される。 The recognition circuit 202 is the ultra-high-speed pattern recognition circuit described above, and the execution command is input to the serial-parallel converter 110, and N light pulses are generated corresponding to the input N-bit input word. . The optical pulses generated by each conversion channel 111 are input to the decision circuit 120 in each decision stage S via an optical waveguide 121 .
 コントローラ201によって生成される電気信号の持続時間は、上述したように、光パルスの約N倍である。例えば、各光パルスが10ビットを40psの持続時間で表すとすると、コントローラ201は、2.5Gb/sのクロックで動作し、一般的な電子回路の速度で動作することができる。光パルスが対応するビット数を増やせば、コントローラ201のクロック速度はさらに緩和され、クロック速度が同じであれば、より高い計算処理能力を得ることができる。 The duration of the electrical signal generated by the controller 201 is approximately N times the duration of the light pulse, as described above. For example, if each optical pulse represents 10 bits with a duration of 40 ps, the controller 201 can clock at 2.5 Gb/s and operate at the speed of common electronic circuits. By increasing the number of bits to which the light pulses correspond, the clock speed of the controller 201 can be further relaxed, resulting in higher computational throughput at the same clock speed.
 認識回路202の出力は、コントローラ201から出力される実行命令の認識結果として、Nビットの組み合わせに一意対応する2N個の出力ポートのいずれかから、制御信号として電気信号を出力する。すなわち、超高速パターン認識回路において、最下位のN番目の決定段階Sにおける2N-1個の決定ユニットUN-s,t(t=2N-1)のうちの1つの決定ユニットからの出力であって、2つの出力ポートのいずれか一方の出力ポートから、最終結果としての電気信号が出力される。認識回路202から出力された電気信号は、メモリ203の光共振器232を駆動し、メモリ203に予め記憶されている計算結果を検索する。メモリ203の詳細は後述する。メモリ203として高速光メモリを例に説明するが、本実施形態のコンピューティングプラットフォームにおいては、通常の電子回路からなるメモリを用いることもできる。 The recognition circuit 202 outputs an electric signal as a control signal from one of 2 N output ports uniquely corresponding to the combination of N bits as a recognition result of the execution instruction output from the controller 201 . That is, in the ultrafast pattern recognition circuit, from one decision unit out of 2 N−1 decision units U N−s,t (t=2 N−1 ) in the lowest N-th decision stage S N , and an electrical signal as a final result is output from one of the two output ports. The electrical signal output from the recognition circuit 202 drives the optical resonator 232 of the memory 203 to retrieve the calculation results pre-stored in the memory 203 . Details of the memory 203 will be described later. A high-speed optical memory will be described as an example of the memory 203, but in the computing platform of this embodiment, a memory made up of ordinary electronic circuits can also be used.
 メモリ203の検索結果は、計算結果を表す一連の直列光パルスとして出力される。直列光パルスは、選択回路204の光スイッチ241により、コントローラ201に送り返されるか、または次の計算ステップの新しいオペランドとして用いられるように、認識回路202に振り分けられる。コントローラ201に送り返される光パルスは、電気信号に変換され、シリアル-パラレル変換器(S/P)242を介して出力される。 The search result of memory 203 is output as a series of serial light pulses representing the calculation result. The serial light pulses are routed by the optical switch 241 of the selection circuit 204 to the recognition circuit 202 to be sent back to the controller 201 or used as new operands in the next computational step. Optical pulses sent back to the controller 201 are converted into electrical signals and output via a serial-to-parallel converter (S/P) 242 .
 図9に、本実施形態のコンピューティングプラットフォームの光メモリの概略の構成を示す。メモリ203は、入力から出力までのすべてを光信号で処理する全光メモリである。メモリ203は、認識回路202の2N個の出力ポートに対応して、2N個の光共振器232を備える(説明の便宜上4つの出力ポートの場合を示す)。複数の光共振器232a-dは、光パルス光源231に接続された光導波路237と光学的に結合されている。認識回路202のいずれか1つの出力ポートから電気信号が出力されて、対応する光共振器の1つが駆動されると、駆動された光共振器は、光パルス光源231からの光パルスを、分割回路233a-dに出力する。分割回路233は、光パルスを多数の光パルスに分割する光経路を有し、それぞれの光経路には、所定の遅延時間に設定された遅延回路236が設定されている。分割された光パルスは、空間光位相変調器であるLCOS(液晶オンシリコン)234により反射または吸収される。反射された光パルスは、元の光経路を伝播し、サーキュレータ235を介して、一連の直列光パルスとして出力される。 FIG. 9 shows a schematic configuration of the optical memory of the computing platform of this embodiment. The memory 203 is an all-optical memory that processes everything from input to output with optical signals. The memory 203 has 2 N optical resonators 232 corresponding to the 2 N output ports of the recognition circuit 202 (a case of four output ports is shown for convenience of explanation). A plurality of optical resonators 232 a - d are optically coupled with an optical waveguide 237 connected to the optical pulse source 231 . When an electrical signal is output from any one output port of the recognition circuit 202 to drive one of the corresponding optical resonators, the driven optical resonator divides the optical pulse from the optical pulse light source 231. output to circuits 233a-d. The splitting circuit 233 has optical paths for splitting an optical pulse into a large number of optical pulses, and each optical path is provided with a delay circuit 236 having a predetermined delay time. The split light pulses are reflected or absorbed by LCOS (liquid crystal on silicon) 234, which is a spatial light phase modulator. The reflected light pulse propagates along the original light path and is output as a series of serial light pulses through circulator 235 .
 分割回路233の遅延回路236の遅延時間と、LCOS234の反射制御とを、予め設定しておくことにより、入力された1つの光パルスから、所望の計算結果を表す一連の直列光パルス(図9においては「10110101」という計算結果を表す光パルス)を導出することができる。認識回路202で認識された実行命令は、メモリ203において光学的に処理されるので、従来の電子回路からなるメモリと比較して、応答時間が短縮され、消費電力を低減することができる。 By presetting the delay time of the delay circuit 236 of the division circuit 233 and the reflection control of the LCOS 234, a series of serial light pulses representing a desired calculation result can be generated from one input light pulse (FIG. 9). can derive a light pulse representing the calculation result of "10110101". Since the execution instruction recognized by the recognition circuit 202 is optically processed in the memory 203, the response time can be shortened and the power consumption can be reduced as compared with a memory made up of a conventional electronic circuit.
 図10に、本実施形態の光メモリの他の構成例を示す。より高い計算処理能力を得るために、プロセッサの処理ビット数を増加させる一例を示す。ビット数の増加により、認識回路の出力ポートはべき乗で増加するので、光メモリにおける分割回路、LCOSの構成も飛躍的に増大する。そこで、メモリ303は、複数の波長の光パルスを用いて、波長分割多重によって処理する。 FIG. 10 shows another configuration example of the optical memory of this embodiment. An example of increasing the number of processing bits of a processor to obtain higher computational throughput is shown. As the number of bits increases, the number of output ports of the recognition circuit increases exponentially, so the configuration of the division circuit and LCOS in the optical memory also increases dramatically. Therefore, the memory 303 processes by wavelength division multiplexing using optical pulses of a plurality of wavelengths.
 光パルス光源331は、複数の波長の光パルスを送出する多波長光源である。例えば、入力ワードには、実行命令に加えて波長選択のための付加ビットを付与しておく。認識回路202は、実行命令の認識結果として、2N個の出力ポートのいずれか1つから最終結果となる電気信号OUTと、波長選択用の付加ビットWSELとを出力する。光パルス光源331は、付加ビットWSELに応じた1つの波長を選択して光パルスを送出し、電気信号OUTが入力された光共振器332は、選択された波長の光パルスを分割回路に出力する。 The optical pulse light source 331 is a multi-wavelength light source that emits optical pulses of multiple wavelengths. For example, an input word is given an additional bit for wavelength selection in addition to an execution command. The recognition circuit 202 outputs an electric signal OUT as a final result and an additional bit WSEL for wavelength selection from one of 2 N output ports as a result of recognizing an execution command. The optical pulse light source 331 selects one wavelength corresponding to the additional bit WSEL and outputs an optical pulse, and the optical resonator 332 to which the electrical signal OUT is input outputs the optical pulse of the selected wavelength to the division circuit. do.
 分割回路の前段には、分波回路としてアレイ導波路格子(AWG)337が挿入されており、波長に応じて、光共振器332から出力された光パルスを分割回路333a-dに分岐する。分割回路においては、分割された光パルスが、LCOS334により反射を制御され、サーキュレータ335を介して、一連の直列光パルスとして出力される。 An arrayed waveguide grating (AWG) 337 is inserted as a branching circuit in the preceding stage of the dividing circuit, and the optical pulse output from the optical resonator 332 is branched to the dividing circuits 333a-333d according to the wavelength. In the splitter circuit, the split light pulses are controlled in reflection by LCOS 334 and output as a series of serial light pulses through circulator 335 .
 本実施形態によれば、従来の電子回路の高集積化によらず、高速動作が可能なコンピューティングプラットフォームを提供することができる。 According to this embodiment, it is possible to provide a computing platform capable of high-speed operation without depending on the conventional high integration of electronic circuits.

Claims (5)

  1.  コントローラから出力されるオペレータとオペランドとを含む実行命令を認識する認識回路であって、前記実行命令のビットごとに対応する決定段階を制御する光パルスを出力するシリアル-パラレル変換器、および前記光パルスにより前記決定段階の出力が選択されて、前記実行命令に一意に対応した出力ポートから制御信号を出力する決定回路を含む認識回路と、
     前記制御信号に応じて予め記憶されている計算結果を出力するメモリと
     を備えたことを特徴とする光電子融合型コンピュータ。
    A recognition circuit for recognizing an execution instruction comprising an operator and an operand output from a controller, the serial-to-parallel converter outputting a light pulse controlling a decision stage corresponding to each bit of said execution instruction, and said light a recognition circuit including a decision circuit for outputting a control signal from an output port uniquely corresponding to the execution instruction when the output of the decision stage is selected by a pulse;
    and a memory for outputting pre-stored calculation results according to the control signal.
  2.  前記メモリから出力された計算結果を、前記コントローラに送信し、または次の実行命令のオペランドとして前記認識回路に送信する選択回路をさらに備えたことを特徴とする請求項1に記載の光電子融合型コンピュータ。 2. The optoelectronic convergence type according to claim 1, further comprising a selection circuit that transmits the calculation result output from the memory to the controller or to the recognition circuit as an operand of the next execution instruction. Computer.
  3.  前記メモリは、
     前記認識回路の前記出力ポートごとに対応した光共振器であって、前記制御信号により光源からの光パルスを分岐する光共振器と、
     前記光共振器から分岐された光パルスを分割する分割回路であって、分割された光経路に予め設定された遅延回路を含む分割回路と、
     前記分割回路の光経路から出力される光パルスの反射を制御する空間光位相変調器とを含み、
     前記計算結果は、前記空間光位相変調器で反射された光パルスが前記分割回路から一連の直列光パルスとして出力されることを特徴とする請求項1または2に記載の光電子融合型コンピュータ。
    The memory is
    an optical resonator corresponding to each of the output ports of the recognition circuit, the optical resonator branching an optical pulse from a light source according to the control signal;
    a splitting circuit for splitting the optical pulse split from the optical resonator, the splitting circuit including a preset delay circuit in the split optical path;
    a spatial optical phase modulator for controlling reflection of optical pulses output from the optical path of the splitting circuit;
    3. The opto-electronic hybrid computer according to claim 1, wherein said calculation result is output from said dividing circuit as a series of serial optical pulses of the optical pulses reflected by said spatial optical phase modulator.
  4.  前記認識回路の前記決定回路は、
     前記光パルスが入力される受光素子と、
     前段の決定段階から出力された信号を、前記受光素子の出力に応じて、2つの出力ポートのうちのいずれかに信号を出力する判定回路とを含むことを特徴とする請求項1、2または3に記載の光電子融合型コンピュータ。
    The decision circuit of the recognition circuit comprises:
    a light receiving element to which the light pulse is input;
    and a determination circuit for outputting a signal output from the preceding determination step to one of two output ports according to the output of the light receiving element. 4. The opto-electronic integrated computer according to 3 above.
  5.  前記光源は、多波長光源であり、
     前記メモリは、前記光共振器から分岐された光パルスを波長に応じて分岐して、分割回路に出力する分波回路をさらに含むことを特徴とする請求項3または4に記載の光電子融合型コンピュータ。
    The light source is a multi-wavelength light source,
    5. The opto-electronic convergence type according to claim 3, wherein the memory further includes a demultiplexing circuit that demultiplexes the optical pulse demultiplexed from the optical resonator according to wavelength and outputs the divided optical pulse to a demultiplexing circuit. Computer.
PCT/JP2021/001839 2021-01-20 2021-01-20 Photonics/electronics integrated computer WO2022157853A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2022576270A JP7495646B2 (en) 2021-01-20 2021-01-20 Photonics-electronics hybrid computer
PCT/JP2021/001839 WO2022157853A1 (en) 2021-01-20 2021-01-20 Photonics/electronics integrated computer

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2021/001839 WO2022157853A1 (en) 2021-01-20 2021-01-20 Photonics/electronics integrated computer

Publications (1)

Publication Number Publication Date
WO2022157853A1 true WO2022157853A1 (en) 2022-07-28

Family

ID=82548558

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2021/001839 WO2022157853A1 (en) 2021-01-20 2021-01-20 Photonics/electronics integrated computer

Country Status (2)

Country Link
JP (1) JP7495646B2 (en)
WO (1) WO2022157853A1 (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS5846384A (en) * 1981-08-29 1983-03-17 アーエーゲー、オリムピア、アクチェンゲゼルシャフト Method of indicating ideographic symbol and similar graphic
JP2006234964A (en) * 2005-02-22 2006-09-07 Nippon Telegr & Teleph Corp <Ntt> Optical resonator element, optical memory using the same, and optical switch
WO2018194115A1 (en) * 2017-04-19 2018-10-25 日本電信電話株式会社 Signal processing circuit, distributed memory in which same is used, rom, and dac

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS5846384A (en) * 1981-08-29 1983-03-17 アーエーゲー、オリムピア、アクチェンゲゼルシャフト Method of indicating ideographic symbol and similar graphic
JP2006234964A (en) * 2005-02-22 2006-09-07 Nippon Telegr & Teleph Corp <Ntt> Optical resonator element, optical memory using the same, and optical switch
WO2018194115A1 (en) * 2017-04-19 2018-10-25 日本電信電話株式会社 Signal processing circuit, distributed memory in which same is used, rom, and dac

Also Published As

Publication number Publication date
JPWO2022157853A1 (en) 2022-07-28
JP7495646B2 (en) 2024-06-05

Similar Documents

Publication Publication Date Title
US11057143B1 (en) Wavelength division multiplexing (WDM)-based and multipath interferometry based optical ternary content addressable memory (TCAM)
US6559989B1 (en) Optical packet header processing apparatus for optical packet switch
EP1132766A2 (en) Optical signal processing device using optical gate
Ramachandran et al. Design and performance analysis of all-optical cascaded adder using SOA-based MZI
US5093565A (en) Apparatus for sequential optical systems where an independently controllable transmission gate is interposed between successive optoelectronic gates
WO2022157853A1 (en) Photonics/electronics integrated computer
US8259376B2 (en) Delayed optical logic gates
US11783878B2 (en) Time division multiplexing (TDM) based optical ternary content addressable memory (TCAM)
JP6701443B2 (en) Signal processing circuit, distributed memory using the same, ROM and DAC
Li et al. Optical look up table
Ying et al. Sequential logic and pipelining in chip-based electronic-photonic digital computing
London et al. WDM ternary content-addressable memory for optical links
JP6699826B2 (en) Optical logic circuit
Geldenhuys et al. All-optical signal processing for optical packet switching
Shao et al. Robust training of optical neural network with practical errors using genetic algorithm: A case study in silicon-on-insulator-based photonic integrated chips
JP6631921B2 (en) Optical computing unit
Frietman et al. An electro-optic data communication system for the Delft parallel processor
Matsuo et al. Space and power reduction in BDD-based optical logic circuits exploiting dual ports
CN113568471B (en) Photoelectric hybrid adder
Li et al. Complementary logic interface for high performan optical computing with OLUT
Hasnain et al. Towards reconfigurable optoelectronic hardware accelerator for reservoir computing
Mourgias-Alexandris Optical neural networks for deep learning applications
RU2119182C1 (en) Optical analog-to-digital converter
JP6468607B2 (en) Optical computing unit
Pei et al. An Electro-Optic Mixed Structure Multiplier with an Optical Delay Accumulation Scheme in Time-Domain

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21920967

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022576270

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21920967

Country of ref document: EP

Kind code of ref document: A1