# US 2023/0068675 A1 · Encoder and operation method thereof > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2023/0068675 A1 - Document ID: US20230068675A1 - Status: Published application - Kind: A1 - Application number: 17/895,532 - Filed: 2022-08-25 - Published: 2023-03-02 - Earliest claimed: 2021-08-26 - Assignees: Electronics and Telecommunications Research Institute - Applicants: Electronics and Telecommunications Research Institute - Inventors: Sung Eun KIM; Tae Wook KANG; Hyuk KIM; Young Hwan BAE; Kyung Jin BYUN; Kwang IL OH; Jae-Jin LEE; In San JEON - Web page: https://patentgrep.com/patent/US20230068675A1 ## CPC - G06N 3/065: Analogue means - G06N 3/049: Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Disclosed is an encoder including event layer outputs first and second event signals, weight layer applies first and second weights to the first and second event signals respectively, and provides the first event signal in which the first weight is applied and the second event signal in which the second weight is applied to first node, and first spike generation circuit generates first input spike signal of which firing period is changed based on voltage level of the first node. The voltage level of the first node is reduced continuously, increases for first voltage corresponding to the first weight in response to the first event signal activated, and increases for second voltage corresponding to the second weight in response to the second event signal activated. ## Claims ### Claim 1 An encoder comprising: an event layer configured to output a first event signal and a second event signal; a weight layer configured to apply a first weight to the first event signal, to apply a second weight to the second event signal, and to provide the first event signal in which the first weight is applied and the second event signal in which the second weight is applied to a first node; and a first spike generation circuit configured to generate a first input spike signal of which a firing period is changed based on a voltage level of the first node, wherein the voltage level of the first node is reduced continuously, increases for a first voltage corresponding to the first weight in response to the first event signal activated, and increases for a second voltage corresponding to the second weight in response to the second event signal activated, wherein, when the voltage level of the first node is a first level, a firing period of the first input spike signal is a first time interval, and wherein, when the voltage level of the first node is a second level lower than the first level, the firing period of the first input spike signal is a second time interval longer than the first time interval. ### Claim 2 (depends on 1) The encoder of claim 1, wherein, when the voltage level of the first node is not greater than a threshold voltage level, the first input spike signal does not fire. ### Claim 3 (depends on 1) The encoder of claim 1, wherein, whether to activate the first event signal is determined based on whether a first event occurs, and wherein whether to activate the second event signal is determined based on whether a second event occurs. ### Claim 4 (depends on 3) The encoder of claim 3, wherein the first input spike signal includes time-series correlation information between the first event and the second event. ### Claim 5 (depends on 3) The encoder of claim 3, further comprising: a second spike generation circuit configured to generate a second input spike signal of which a firing period is changed based on a voltage level of a second node, wherein the event layer further outputs a third event signal, whose activation is determined based on whether the first event occurs, and a fourth event signal, whose activation is determined based on whether the second event occurs, wherein the weight layer applies a third weight to the third event signal, applies a fourth weight to the fourth event signal, and provides the third event signal in which the third weight is applied and the fourth event signal in which the fourth weight is applied to the second node, and wherein the voltage level of the second node is reduced continuously, increases for a third voltage corresponding to the third weight in response to the third event signal activated, and increases for a fourth voltage corresponding to the fourth weight in response to the fourth event signal activated. ### Claim 6 (depends on 1) The encoder of claim 1, further comprising: a discharge circuit connected to the first node and configured to continuously reduce the voltage level of the first node. ### Claim 7 (depends on 1) The encoder of claim 1, wherein the voltage level of the first node is continuously reduced by charge leakage. ### Claim 8 (depends on 1) The encoder of claim 1, further comprising: a capacitor connected between the first node and a ground voltage. ### Claim 9 (depends on 1) The encoder of claim 1, wherein the first spike generation circuit includes a voltage controlled oscillator (VCO). ### Claim 10 (depends on 1) The encoder of claim 1, wherein the first input spike signal is provided to a spike neural network (SNN) circuit. ### Claim 11 (depends on 10) The encoder of claim 10, wherein the SNN circuit is implemented by a rate-coding method. ### Claim 12 A method of operating an encoder configured to provide an input spike signal to an spike neural network (SNN) circuit, the method comprising: generating a plurality of event signals, whose activations are determined based on a plurality of events occur, respectively; applying a weight to each of the plurality of event signals; accumulating a voltage at a first node, at which a voltage level is continuously reduced, based on a weight applied to an activated event signal from among the plurality of event signals; and outputting the input spike signal of which a firing period is changed based on the voltage level of the first node, wherein the firing period is a first time interval when the voltage level of the first node is a first level, and is a second time interval longer than the first time interval when the voltage level of the first node is a second level lower than the first level. ### Claim 13 (depends on 12) The method of claim 12, wherein, when the voltage level of the first node is not greater than a threshold voltage level, the input spike signal does not fire. ### Claim 14 (depends on 12) The method of claim 12, wherein the encoder further includes: a discharge circuit implemented to reduce the voltage level of the first node. ### Claim 15 (depends on 12) The method of claim 12, wherein the SNN circuit is implemented in a rate-coding method. ### Claim 16 A neuromorphic system comprising: an spike neural network (SNN) circuit implemented by a rate-coding method; and an encoder configured to provide an input spike signal to the SNN circuit, wherein the encoder includes a first node, and includes a spike generation circuit configured to generate the input spike signal based on a voltage level of the first node, wherein the voltage level of the first node is continuously reduced, increases by a first voltage when a first event occurs, and increases by a second voltage when a second event occurs, and wherein a firing period of the input spike signal is reduced when the voltage level of the first node increases. ### Claim 17 (depends on 16) The neuromorphic system of claim 16, wherein the encoder further includes a discharge circuit connected to the first node and configured to reduce the voltage level of the first node. ### Claim 18 (depends on 16) The neuromorphic system of claim 16, wherein, when the voltage level of the first node is not greater than a threshold voltage level, the input spike signal does not fire. ### Claim 19 (depends on 16) The neuromorphic system of claim 16, wherein the input spike signal includes occurrence timing information about the first event and the second event.