A multi-level state regulation method for an all-optical synapse device based on PAWM modulation

Abstract

This application belongs to the field of biomimetic synaptic device control, specifically disclosing a multi-level state control method for all-optical synaptic devices based on PAWM modulation. This application uses a crystallization threshold light pulse as the control reference, and then constructs a real-time feedback closed loop: after each control pulse is applied, the change in optical response is immediately detected and compared with a preset range; if it deviates, the energy of subsequent pulses is adaptively adjusted based on PAWM technology. This closed-loop mechanism can correct the nonlinearity and randomness in the phase transition process in real time, ensuring that the step size of each state change is controllable and uniform. By cyclically executing this process, the optical state can gradually and stably transform through a large number of controlled fine-tunings. Compared with static or open-loop adjustment, high-precision linear control of the crystallization process is achieved, thereby significantly increasing the number of intermediate states, fundamentally improving the state resolution of the device, and realizing a leap from finite states to multi-level, approximately continuously adjustable states.

Description

Technical Field

[0001] This application belongs to the field of biomimetic synaptic device control, specifically involving a multi-level state control method for an all-optical synaptic device based on PAWM modulation. Background Technology

[0002] In neuromorphic computing architectures, hardware synapses are the core devices for realizing information transmission and weight plasticity between neurons. In recent years, all-optical synaptic devices based on phase-change materials (PCMs) have become strong candidates for simulating synaptic function due to their non-volatility, multi-state storage capabilities, and high-speed optical modulation characteristics. These PCM-based all-optical synaptic devices utilize light pulses to induce reversible transitions between amorphous and crystalline states, achieving light intensity modulation and thus simulating the weight modulation (i.e., connection strength) process of biological synapses. Compared to electronic synapses, these devices can directly perform signal processing and storage in the optical domain, significantly improving information transmission speed and bandwidth while reducing energy consumption, providing a feasible path for constructing large-scale, high-speed, and low-power neuromorphic optical computing systems.

[0003] The fine-tuning of synaptic weights directly determines the learning ability and memory accuracy of a neural network. The more states a device has, the closer it approximates the continuous plasticity of biological synapses, enabling it to store more information, perform more complex calculations, and reduce energy consumption while maintaining high speed. Therefore, multi-level adjustable states not only determine the storage capacity of hardware synapses in terms of the number of states, but also the symmetry, linearity, energy efficiency, and dynamic range during state switching in the tuning process are core indicators for evaluating device performance and are key to achieving high-precision, low-power, and high-convergence-speed efficient brain-like optical computing.

[0004] However, existing control methods mainly intervene in the crystallization degree of phase change materials through pulse amplitude modulation, pulse width modulation, or the accumulation of the number of pulses with fixed parameters, thereby adjusting the synaptic weights. However, these schemes reveal significant drawbacks in practical applications: First, the control resolution is limited. Due to the randomness and nonlinearity of the phase change material crystallization process, the device struggles to maintain a large number of stable discrete energy levels, failing to meet the requirements of high-precision simulation calculations. Second, the nonlinearity of weight updates is severe. Because the movement of the crystallization front and energy absorption during the phase change are not in a nonlinear proportion, the step size of the weight changes with the pulse is extremely uneven, increasing the difficulty of algorithm convergence. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this application is to provide a multi-level state control method for all-optical synaptic devices based on pulse-amplitude-width modulation (PAWM) modulation, aiming to solve the problem that all-optical synaptic devices can only achieve a limited number of adjustable states and have low state resolution due to limitations in static adjustment methods.

[0006] The first aspect of this application relates to a multi-level state control method for an all-optical synaptic device based on PAWM modulation, comprising: step S10, generating an initial light pulse based on preset initial energy parameters, and determining a crystallization threshold light pulse that drives the phase change material layer to undergo a crystallization transition by adjusting the initial light pulse; step S20, applying the current crystallization threshold light pulse to the phase change material layer of the all-optical synaptic device to obtain the change in the optical response of the all-optical synaptic device; step S30, comparing the change in the optical response with a preset change range, and adjusting the energy of the crystallization threshold light pulse based on PAWM when the change in the optical response is outside the preset change range; and step S40, repeating steps S20 and S30 until the optical state of the phase change material transitions to the target state through multiple intermediate states, thereby achieving multi-level state control.

[0007] In one embodiment, step S10 includes: generating an initial light pulse based on a preset initial energy parameter, and generating a pulse sequence based on the initial light pulse that includes multiple light pulses with increasing power from the initial light pulse as a reference; applying the pulse sequence to the phase change material layer of the all-optical synapse device, and obtaining the change in optical response of the all-optical synapse device after each application of a light pulse; when the change in optical response changes, identifying the current light pulse as a crystallization threshold light pulse, and stopping the application of subsequent light pulses.

[0008] In one embodiment, the preset variation range includes a first threshold and a second threshold, wherein the first threshold is less than the second threshold; step S30 includes: if the optical response variation is less than the first threshold, adjusting the parameters of the crystallization threshold light pulse based on PAWM to increase the pulse energy; if the optical response variation is greater than the second threshold, adjusting the parameters of the crystallization threshold light pulse based on PAWM to decrease the pulse energy.

[0009] In one embodiment, adjusting the parameters of the crystallization threshold light pulse based on PAWM to increase the pulse energy specifically involves: decreasing the pulse width with a first preset gradient and increasing the pulse amplitude with a second preset gradient; adjusting the parameters of the crystallization threshold light pulse based on PAWM to decrease the pulse energy specifically involves: increasing the pulse width with a first preset gradient and decreasing the pulse amplitude with a second preset gradient.

[0010] In one embodiment, the change in optical response is a change in transmittance; obtaining the change in optical response specifically involves: measuring the output optical power of a probe beam after passing through the phase change material layer before and after applying an optical pulse to the phase change material layer; and determining the change in transmittance based on the change in output optical power before and after applying the optical pulse.

[0011] In one embodiment, step S40 includes: repeatedly executing steps S20 and S30 to obtain the intermediate state of the all-optical synaptic device in each cycle; when the optical response of the current intermediate state meets the preset target value, the current intermediate state is identified as the target state, and the cycle is stopped.

[0012] The second aspect of this application relates to a control system, comprising: a pulse generation module for generating an initial light pulse based on a preset initial energy parameter, and determining a crystallization threshold light pulse that drives a phase change material layer to undergo a crystallization transition by adjusting the initial light pulse; an execution and detection module for applying the crystallization threshold light pulse to the phase change material layer of an all-optical synapse device and acquiring the optical response change of the all-optical synapse device; a comparison and adjustment module for comparing the optical response change with a preset change range, and adjusting the energy of the crystallization threshold light pulse based on PAWM when the optical response change is outside the preset change range; and a control module for controlling the execution and detection module and the comparison and adjustment module to execute cyclically until the optical state of the phase change material transitions to a target state through multiple intermediate states, thereby achieving multi-level state regulation.

[0013] Thirdly, this application provides a drive control device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0015] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0016] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0017] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: This application effectively solves the existing problems by adopting a closed-loop control method based on real-time feedback and dynamic adjustment. Specifically, firstly, the minimum energy required to drive the phase transition, i.e., the crystallization threshold light pulse, is dynamically determined in step S10, rather than relying on preset fixed parameters. This establishes an accurate starting benchmark for subsequent fine control and avoids directly entering a saturation or insufficient response state due to improper initial energy setting.

[0018] Based on this, steps S20 and S30 constitute the core feedback control loop: after each application of a control pulse, the change in the device's optical response is immediately acquired and compared with a preset ideal range; if the response deviates from the expected range, the energy of the next pulse is immediately adaptively adjusted based on PAWM technology. This closed-loop mechanism of application, detection, comparison, and adjustment ensures that each state change is a precise correction based on an evaluation of the previous control effect, thereby overcoming the nonlinearity and randomness in the phase change material response process and ensuring that the step size of each state switch is controllable and uniform.

[0019] Finally, by repeatedly executing this feedback loop in step S40, the optical state of the phase change material can stably and progressively transition from one state to the target state through a large number of such controlled fine-tuning steps. Compared with existing static or open-loop adjustment techniques, this scheme, by introducing this dynamic adaptive feedback control mechanism, achieves high-precision, linear control of the crystallization process of the phase change material, thereby significantly increasing the number of distinguishable and maintainable stable intermediate states of the device, that is, greatly improving the state resolution, and realizing the leap from finite discrete states to approximately continuous, multi-level adjustable states. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the phase-change all-optical synapse device structure provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the multi-level state control method for an all-optical synaptic device based on PAWM modulation provided in the embodiments of this application. Figure 3 This is a flowchart illustrating the specific implementation of the multi-level state control method for all-optical synaptic devices based on PAWM modulation provided in this application embodiment; Figure 4 This is a finely detailed multi-level control diagram of 348 different states provided in the embodiments of this application; Figure 5 This is provided by the embodiments of this application. Figure 4 A magnified view of a portion of the image; Figure 6This is a histogram of weight adjustment step size distribution provided in the embodiments of this application; Figure 7 This is a structural block diagram of the drive and control system provided in the embodiments of this application; Figure 8 This is a schematic diagram of the drive and control device provided in the embodiments of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.

[0023] In this application, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0024] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0025] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0026] Currently, in neuromorphic computing architectures, hardware synapses are the core components for realizing information transmission and weight plasticity between neurons. The fine-tuning of synaptic weights directly determines the learning ability and memory accuracy of the neural network. The more states a device has, the closer it approximates the continuous plasticity of biological synapses, enabling it to store more information, perform more complex calculations, and reduce energy consumption while maintaining high speed. Therefore, multi-level adjustable states not only determine the storage capacity of hardware synapses in terms of the number of states, but also the symmetry, linearity, energy efficiency, and dynamic range during state switching in the control process are core indicators for evaluating device performance and are key to achieving high-precision, low-power, and high-convergence-speed efficient neuromorphic optical computing.

[0027] The phase-change all-optical synaptic device addressed in this application refers to the number of different states that the device can stably maintain. Each state corresponds to a different optical transmittance. This multi-state modulation reflects the plasticity of the synaptic device and is used to simulate the weight modulation between neurons. The tunable state number test mainly involves adjusting the phase state of the phase change material by controlling the parameters of the input light pulse, such as pulse width, intensity, and frequency, thereby achieving different optical transmittances to simulate the multi-level weight changes of biological synapses.

[0028] It should be noted that, please refer to Figure 1 , Figure 1 This is a schematic diagram of the all-optical synaptic device provided in the embodiments of this application. By inputting pump light pulses of different energies or intensities into the phase change material, the phase change material can be gradually driven to transform between an amorphous state and a crystalline state. By adjusting the intensity, pulse width, and frequency of the pump light, the partial crystallization or partial amorphization of the phase change material can be controlled, thereby generating different phase states, which directly affect the transmittance of the input light signal. By precisely adjusting the pulse parameters, multiple stable intermediate states can be generated, each state corresponding to a partial crystalline state, thus enabling the device to exhibit different light transmittances and forming multiple adjustable weighted states.

[0029] It is understandable that this optical property of phase change materials can be used to control the degree of loss of optical signals during transmission, and different degrees of crystallization of phase change materials can be achieved by finely adjusting different pulse parameters, thereby realizing multi-level control of optical output power. Each level represents a state, and the more adjustable states there are, the better the performance of the phase change synapse device.

[0030] Based on this, this application proposes a multi-level state control method for all-optical synaptic devices based on PAWM modulation. Please refer to... Figure 2 , Figure 2 This is a flowchart illustrating the multi-level state control method for an all-optical synaptic device based on PAWM modulation provided in the embodiments of this application.

[0031] In this embodiment, the above method includes steps S10 to S40.

[0032] Step S10: Based on the preset initial energy parameters, an initial light pulse is generated, and the crystallization threshold light pulse that drives the phase change material layer to undergo crystallization transformation is determined by adjusting the initial light pulse.

[0033] It should be noted that step S10 involves determining the crystallization threshold light pulse that drives the phase change material to undergo a crystallization transition. The initial energy parameter refers to a pre-set combination of physical parameters used to generate the tentative light pulse, typically including the center wavelength, the pulse amplitude characterizing the peak power, and the pulse width characterizing the duration. The initial light pulse is the light signal generated based on this set of parameters.

[0034] Understandably, this step is essentially a calibration or standardization process: by applying a series of initial light pulses with increasing or decreasing energy, determined by both amplitude and pulse width, and simultaneously monitoring the device's optical response, such as transmitted light intensity, the critical light pulse at which the phase change material begins to undergo significant crystallization—that is, when a detectable and irreversible change in the optical response begins—can be found. This pulse is defined as the crystallization threshold light pulse. This threshold light pulse serves as the reference point for subsequent precise multi-level control.

[0035] In actual device operation, this step can be achieved by using a programmable light source, such as a tunable laser, in conjunction with an electro-optic modulator and a high-speed photodetector in conjunction with a data acquisition system, and automatically finding the threshold through a feedback loop.

[0036] In one feasible implementation, step S10 includes: generating an initial light pulse based on a preset initial energy parameter, and generating a pulse sequence based on the initial light pulse that includes multiple light pulses with increasing power from the initial light pulse as a reference; applying the pulse sequence to the phase change material layer of the all-optical synapse device, and obtaining the change in optical response of the all-optical synapse device after each application of a light pulse; when the change in optical response changes, identifying the current light pulse as a crystallization threshold light pulse, and stopping the application of subsequent light pulses.

[0037] Understandably, the core of this method lies in automatically and rapidly detecting the critical energy point that triggers the initial crystallization of the phase change material through a sequence of pulses with progressively increasing power. Preset initial energy parameters serve as the baseline for initiating the process, typically including a low initial peak power and a fixed pulse width and shape. The initial optical pulse is then the first detection pulse generated based on these parameters. The pulse sequence refers to a series of optical pulses generated sequentially with the power of this initial pulse as a baseline, following a preset step size, for example, increasing the peak power by 5% each time. Other parameters of these pulses, such as width, frequency, and wavelength, are usually kept constant. This sequence can be generated by a programmable arbitrary waveform generator controlling a laser driver or an electro-optic modulator.

[0038] It should be noted that applying this pulse sequence to the phase change material layer of the device is a sequential process. After each pulse in the sequence is applied, the change in the device's optical response must be acquired immediately or subsequently. This is typically achieved using a separate, low-power continuous probe beam with a wavelength chosen to provide a significant optical contrast between the amorphous and crystalline states of the phase change material, such as around 1550 nm in communication wavelengths. The intensity of the transmitted or reflected probe beam is measured using a high-speed photodetector, and the intensity after the pulse is compared to the baseline intensity before the pulse. The difference or relative rate of change is the change in the optical response. In the initial stage, when the applied pulse energy is insufficient to trigger a phase transition, this change should be zero or at the system noise level.

[0039] Understandably, the logic for identifying a given optical pulse as a crystallization threshold pulse when the optical response changes is based on the physical properties of the phase change material. When the energy of the optical pulse first reaches or exceeds the material's crystallization energy threshold, the material undergoes initial crystallization in its microstructure, resulting in a measurable change in its refractive index and extinction coefficient. This causes a clear jump in the transmittance or reflectance of the probe light, exceeding the noise range. Detecting this first non-zero change signifies the start of a phase transition in the material, and the peak power of the applied optical pulse at this point is defined as the energy parameter of the crystallization threshold pulse.

[0040] Understandably, once this threshold is detected, the process immediately stops applying subsequent light pulses to prevent unnecessary over-crystallization of the material and ensure that the material is at a known starting point close to the initial amorphous state. This is crucial for subsequent reversible, multi-level precision control.

[0041] In practical implementation, this process can be accomplished through an integrated testing system. This system includes: a tunable laser source and optical modulator for generating pump pulse sequences; a laser diode for generating continuous probe light; an optical coupler for coupling the two beams into an optical waveguide integrated with phase-change material; a high-speed photodetector for measuring the output light intensity; and a main control unit, such as an FPGA or a computer running dedicated software, responsible for controlling pulse sequence generation, data acquisition, real-time comparison, and issuing stop commands.

[0042] Alternative options for this implementation include, but are not limited to: the decision condition does not have to be strictly non-zero; instead, a positive number slightly higher than the system noise level can be set as the decision threshold to characterize this change and improve anti-interference capability. Besides fixed pulse width scanning power, fixed power scanning pulse width can also be used, or a two-dimensional parameter scan of power and pulse width can be performed to more accurately determine the energy threshold surface, but this will significantly increase calibration time. The power increment of the pulse sequence can be adaptive instead of using a fixed step size, for example, dynamically adjusting the power increment of the next step based on the response gradient of previous measurements to accelerate the search speed.

[0043] Step S20: Apply the current crystallization threshold light pulse to the phase change material layer of the all-optical synapse device to obtain the change in the optical response of the all-optical synapse device.

[0044] Step S30: Based on the comparison between the optical response change and a preset change range, and when the optical response change is outside the preset change range, adjust the energy of the crystallization threshold light pulse based on PAWM.

[0045] It should be noted that steps S20 and S30 constitute a core feedback control loop. The current crystallization threshold light pulse refers to the light pulse applied after dynamic adjustment during the control process. This pulse is applied to the loss-modulated phase change material layer of the device. This type of phase change functional layer is usually made of phase change media such as chalcogenides and integrated into an optical waveguide or microcavity structure.

[0046] The change in optical response refers to the change in the device's response to a continuous or low-power probe light after the application of a single or series of pulses. Specifically, it can be quantified as the absolute or relative value of the change in transmittance, reflectance, or phase. The wavelength of the probe light is located in the wavelength range sensitive to the optical properties of the phase change material.

[0047] Taking transmittance as an example, the specific method for obtaining the change in optical response is as follows: before and after applying an optical pulse to the phase change material layer, the output optical power of a probe beam after passing through the phase change material layer is measured respectively; based on the change in output optical power before and after applying the optical pulse, the change in transmittance is determined.

[0048] It should be noted that the preset change range is a pre-defined, narrow target change interval, representing the expected change in response caused by each adjustment step. Therefore, the logic of step S30 is to compare the measured actual change in optical response with the preset change range. If it falls outside this range, it indicates that the applied pulse energy is insufficient or excessive, and adjustment is required.

[0049] It should be noted that PAWM-based energy adjustment refers to precisely controlling the total energy injected into the phase change material by adjusting the amplitude, width, or a combination thereof of subsequent pulses, based on the principle that energy ≈ peak power × pulse width, thereby controlling the degree of crystallization or amorphization. For example, if the change is less than the lower limit of the preset range, the pulse amplitude or width needs to be increased by a certain proportion to provide more energy; if the change exceeds the upper limit, the energy needs to be reduced.

[0050] In one feasible implementation, the preset change range can be a pre-defined quality control interval used to evaluate whether the effect of a single adjustment meets the target. This range is jointly defined by a first threshold and a second threshold, where the first threshold is the lower limit of the range, the second threshold is the upper limit of the range, and the first threshold is less than the second threshold. This range represents the ideal increment of the change in the device's optical response, such as the change in transmittance, after a single light pulse is applied. For example, if the goal is to eventually increase the transmittance from 10% to 80%, and it is planned to be completed in 100 steps, then the ideal change in each step is approximately 0.7%. The preset change range can be set to 0.6% to 0.8%, with a first threshold of 0.6% and a second threshold of 0.8%, to ensure that each step is neither too small nor too large.

[0051] The specific control logic is as follows: After applying a current crystallization threshold light pulse and measuring the change in optical response, this measured value is compared with a preset range. If the measured change is less than the first threshold, i.e., below the lower limit of the acceptable range, it indicates that the energy of the injected phase change material is insufficient and fails to drive the material to crystallize to a sufficient degree. In this case, the parameters of the crystallization threshold light pulse need to be adjusted based on PAWM to increase the pulse energy. Conversely, if the measured change is greater than the second threshold, i.e., exceeding the upper limit of the acceptable range, it indicates that the injected energy is excessive, leading to excessive crystallization changes. In this case, the parameters need to be adjusted based on PAWM to reduce the pulse energy.

[0052] It should be noted that this reduction in energy does not refer to a return to the crystallization state, but rather that the energy falls into a suitable range, causing the change in optical response to fall between the first and second thresholds, thus keeping the growth within a suitable range.

[0053] Understandably, in the process of multi-step approximation from a certain state to the target state, each step aims to produce a moderate and controlled change in the device's optical response. If the measured change is less than a first threshold, the energy of the next pulse needs to be increased to drive a more significant phase transition in the material; if the measured change is greater than a second threshold, the energy of the next pulse needs to be decreased to prevent excessive phase transition. This energy reduction operation is specific to the current control step and is intended to correct for excessive changes in the current step, ensuring that the change in the next step falls back to the ideal range. Essentially, it is a dynamic calibration of the control step precision, rather than a directional reversal of the overall crystallization process. Through this continuous and adaptive energy fine-tuning, it can be ensured that the entire transition from the initial state to the target state is smooth, precise, and controllable, with each step's progress within an appropriate range.

[0054] The energy can be adjusted in the following ways: Adjusting the parameters of the crystallization threshold light pulse based on PAWM to increase the pulse energy specifically involves: decreasing the pulse width with a first preset gradient and increasing the pulse amplitude with a second preset gradient; adjusting the parameters of the crystallization threshold light pulse based on PAWM to decrease the pulse energy specifically involves: increasing the pulse width with a first preset gradient and decreasing the pulse amplitude with a second preset gradient.

[0055] Understandably, to increase pulse energy, the pulse width is reduced by a first preset gradient, while the pulse amplitude is increased by a second preset gradient. This operation shortens the pulse duration while increasing the instantaneous power of the pulse. The combined effect is that while the total injected energy increases due to the increased amplitude, the energy is injected with higher peak power in a shorter time due to the shortened pulse width. This high-power short pulse is beneficial for rapidly establishing the required thermal field in phase change materials, promoting the formation and growth of crystal nuclei. At the same time, the shorter duration helps to limit the range of thermal diffusion and improve the spatial locality of control, which is crucial for avoiding thermal crosstalk in highly integrated devices.

[0056] Understandably, to reduce pulse energy, the pulse width is increased with a first preset gradient and the pulse amplitude is decreased with a second preset gradient. This operation prolongs the pulse duration while reducing the instantaneous power of the pulse. The combined effect is that the total injected energy decreases due to the reduced amplitude, and the decrease in peak power makes the energy injection rate more gradual. This low-power, longer pulse width is beneficial for achieving more uniform and controllable energy deposition when fine control and avoiding excessive phase transition are required, thereby precisely limiting the degree of phase transition within the desired small increment.

[0057] It should be noted that the first preset gradient specifically refers to the fixed or programmable unit of change used when adjusting the pulse width parameter. It is a value with time dimensions, such as 1 nanosecond or a percentage relative to the current pulse width. When the pulse width needs to be adjusted, the control unit will add or subtract the value represented by the first preset gradient from the current pulse width value, thereby generating a pulse with a new pulse width.

[0058] It should be noted that the second preset gradient specifically refers to the fixed or programmable unit of change used when adjusting the pulse amplitude parameter. It is a value with power or intensity dimensions, such as 0.02 watts or a percentage relative to the current amplitude. When amplitude adjustment is needed, the control unit adds or subtracts the value represented by the second preset gradient from the current amplitude, thereby generating a pulse with a new amplitude.

[0059] Understandably, the first / second preset gradients used to increase energy and the gradient values ​​used to decrease energy can be set to different magnitudes to accommodate the asymmetric kinetics of the crystallization and amorphization processes. Furthermore, a fixed mathematical relationship, such as a proportional relationship, can exist between the first and second preset gradients, ensuring that pulse width and amplitude adjustments are always made at a specific ratio to maintain a certain pulse shape characteristic. Alternatively, the gradient values ​​may not be fixed but obtained by looking up a table or calculating based on the current intermediate state of the device to compensate for the nonlinearity of the phase transition process.

[0060] Step S40: Repeat steps S20 and S30 until the optical state of the phase change material transitions to the target state through multiple intermediate states, thereby achieving multi-level state adjustment.

[0061] Understandably, step S40 describes the gradual and controllable transition from one state to another target state through iterative feedback control loops. Repeated execution means applying the sequence of application, measurement, comparison, and adjustment of steps S20 and S30 multiple times. Multiple intermediate states refer to a series of metastable states generated between a completely amorphous and a completely crystalline state by controlling the different crystallization ratios of the phase change material. Each intermediate state has a unique and stable optical constant, corresponding to a specific synaptic weight value.

[0062] Understandably, the target state is a pre-defined, desired specific optical response (weighted) state. Through PAWM modulation, each iteration drives only a small, controlled phase transition in the material, thereby gradually and accurately approaching the target state, ultimately achieving high-precision, multi-level state control.

[0063] In one feasible implementation, step S40 includes: repeatedly executing steps S20 and S30 to obtain the intermediate state of the all-optical synaptic device in each cycle; when the optical response of the current intermediate state meets the preset target value, the current intermediate state is identified as the target state, and the cycle is stopped.

[0064] It is understood that this embodiment explicitly states that the transition from the initial state to the target state is accomplished through an iterative feedback loop, and defines the precise conditions for loop termination. Repeated execution of steps S20 and S30 constitutes a complete control cycle. Upon completion of each such cycle, the optical response of the phase change material, such as transmittance, undergoes a controlled change relative to the previous cycle, thereby entering a new, stable intermediate state. Each intermediate state corresponds to a specific partial crystallization ratio of the phase change material and possesses unique and reproducible optical constants.

[0065] It's important to note that the core control logic of this process involves the system acquiring not only the change in optical response generated by step S20 (i.e., the single-step increment) after each loop, but more importantly, acquiring and recording the optical response value of the current intermediate state. This optical response value refers to the total optical response value accumulated from an absolute reference, such as the transmitted light intensity in a completely amorphous state, to the present, for example, the current transmittance. The system compares this accumulated current optical response value with a preset target value set by the user or the upper-level algorithm. This preset target value represents the optical response value corresponding to the specific weight that the synaptic device is expected to ultimately achieve.

[0066] It should be noted that when the error between the current measured value and the target value falls within an acceptable tolerance range, the current intermediate state is considered the target state. Subsequently, the system stops looping and no new control pulses are applied. This marks the successful completion of a multi-level, step-by-step control process from the initial state to the target state. This method ensures that the endpoint of the control is precise and predictable, rather than depending on a fixed number of pulses.

[0067] Understandably, this method achieves precise control over the crystallization / amorphization process of phase change materials by jointly modulating the amplitude and width of the optical pulse, thereby generating and stably maintaining a series of intermediate states with different optical properties between the amorphous and crystalline states, that is, realizing multi-level and reversible adjustment of synaptic weights.

[0068] Based on the above embodiments, this application provides a specific implementation method. Please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the specific implementation of the multi-level state control method for all-optical synaptic devices based on PAWM modulation provided in this application.

[0069] STEP 1 (Setting Initial Parameters): The entire process begins by setting a suitable set of initial optical pulse parameters. As shown in the example, the initial parameters can be set to a pulse width of 100 nanoseconds and a power of 2.5 milliwatts. The purpose of this step is to establish a reliable starting point for the subsequent search and control process. The initial optical pulse is generated based on this set of parameters. The selection of these initial values ​​is usually based on prior knowledge of the fundamental properties of the phase change material, ensuring that the initial pulse energy is below the crystallization threshold, thereby guaranteeing that the initial state is a defined amorphous state.

[0070] STEP 2 (Finding the Crystallization Threshold Pulse): This step calibrates the crystallization threshold. Specifically, the pulse width is fixed, for example, maintained at 100 nanoseconds as set in STEP 1, and then the pulse power is gradually and monotonically increased. After each pulse is applied, the change in the device's optical response is monitored in real time, denoted as ΔT in the figure. This embodiment uses the change in transmittance as an example for explanation, with the unit being decibels (dB). When a detectable and definite change in ΔT from zero or the noise baseline is detected for the first time (e.g., ΔT > 0), the currently applied pulse is determined to be the crystallization threshold pulse. This pulse is the minimum energy pulse capable of inducing initial crystallization in the phase change material, establishing an energy reference for subsequent fine multi-step control. The "Intermediate state" illustration in the figure indicates that this step transitions the device from a completely amorphous state to the first distinguishable intermediate state.

[0071] STEP 3 (Judgment and PAWM Feedback Control): This is the core step, forming a feedback loop. After obtaining the crystallization threshold pulse, the process enters the loop control stage: First, ΔT is judged by measuring the change in optical response ΔT after each application of the current parameter pulse. Then, branch control is performed based on preset ranges, with the first threshold set at 0.1 dB and the second threshold set at 0.2 dB.

[0072] Specifically, if ΔT < 0.1 dB, it indicates that the current state change is insufficient. In this case, the system initiates PAWM (Power-On-Demand) to increase energy. According to the method of this application, this is typically achieved by reducing the pulse width and simultaneously increasing the pulse power, thereby increasing the peak power while improving the total energy and promoting more significant crystallization.

[0073] Specifically, if 0.1 dB ≤ ΔT ≤ 0.2 dB, it indicates that the change in state is within the ideal preset range. In this case, the system maintains the parameters, that is, keeps the pulse width and power of the current pulse unchanged, and repeats this parameter in the next step to achieve stable and consistent stepping.

[0074] Specifically, if ΔT > 0.2 dB, it indicates that the current state change is too large, posing a risk of overshoot. In this case, the system initiates PAWM (Power-On-Wave) energy reduction operation. According to the method of this application, this is typically achieved by increasing the pulse width and simultaneously decreasing the pulse power, thereby reducing the peak power while reducing the total energy, ensuring that the next change is more precise and controlled.

[0075] Finally, after completing the above judgment and operation once, the process moves to the next step, repeating STEP3, forming a closed-loop iteration. This cycle continues, with each step causing a controlled, small change in the crystallinity of the device, i.e., its state, until the material reaches crystallization saturation, the control reaches its limit, and the phase change material is nearly fully crystallized, meaning that even if a pulse is applied again, ΔT will no longer change significantly. At this point, the process terminates, and the total number of states is output. The total number of states equals the sum of the number of all stable, distinguishable intermediate states between the initial amorphous state and the final saturated crystallized state. This value is a key indicator for measuring the plasticity performance of the synaptic device, such as the accuracy of weight expression and storage capacity.

[0076] Specifically, please refer to Figure 4 , Figure 4 This is a finely detailed multi-level control diagram of 348 different states provided in the embodiments of this application.

[0077] Understandably, according to the above process, Figure 4 The continuously adjustable range from high transmittance (approximately 0 dB) to low transmittance (approximately -29.5 dB) shown is a direct result of the successful execution of the modulation method described in this application. Through amplitude and pulse width co-modulation, multi-step low-energy accumulation, and dynamic feedback closed loop, the change in optical response caused by each state switch can be precisely controlled within a preset range (e.g., 0.1 dB to 0.2 dB). A total modulation depth of 29.5 dB is achieved by dividing the modulation into 348 stable states, meaning that the average transmittance step change between each state is approximately 0.085 dB.

[0078] Understandably, the continuous state region shown by the blue band in the figure is accumulated through a large number of such controlled micro-steps, exhibiting good linearity. Meanwhile, the total modulation depth of 29.5 dB, divided into 348 stable states, exceeds 8 bits, meaning there are more than 256 adjustable states.

[0079] Further, please refer to Figure 5 , Figure 5 This is provided by the embodiments of this application. Figure 4 A magnified view of a portion thereof. This application also shows the parameter record tables for states 50 to 82, within the area highlighted in red.

[0080] Table 1:

[0081] Understandably, a transmittance change of 0.05~0.08 dB is considered a distinguishable state. When the change is less than 0.05 dB, the pulse width is reduced while the pulse amplitude is increased, resulting in an overall pulse energy greater than the previous pulse. When the change is greater than 0.08 dB, the pulse width is increased while the pulse amplitude is decreased, resulting in an overall pulse energy less than the previous pulse, thereby achieving continuous control.

[0082] Please refer to Figure 6 , Figure 6 This is the weight adjustment step size distribution histogram provided in the embodiments of this application. Figure 6 This paper presents the statistical distribution of all single transmittance changes, i.e., weighted adjustment step sizes, recorded during the continuous control process from amorphous to crystalline state using the aforementioned multi-level state control method of this application. This distribution visually verifies the effectiveness of the control method in achieving high-precision, high-consistency state programming.

[0083] Specifically, the horizontal axis of this histogram represents the transmittance change in dB, indicating the change in device transmittance after each applied modulation light pulse, ranging from 0.00 dB to 0.18 dB. The vertical axis represents the number of states, indicating the frequency of occurrence within the corresponding range of change. A series of light blue bars constitute the main distribution, and a smooth dark blue curve outlines its overall trend. A red vertical dashed line, labeled as system noise fluctuation ±0.02 dB, is placed in the histogram as a key reference benchmark.

[0084] The distribution pattern shows that the vast majority of adjustment step lengths are concentrated in the range of 0.05 dB to 0.12 dB, exhibiting an approximately normal distribution, with the peak occurring around 0.06 to 0.08 dB. Key statistics marked in the upper right corner quantitatively confirm this observation: the mean of all step lengths is 0.085 dB, and the standard deviation is 0.031 dB.

[0085] Understandably, the mean step size is a small and controllable value, which directly proves that this application can achieve fine optical response control, thereby supporting the generation of a large number of distinguishable states, such as the 348 mentioned above. The small standard deviation indicates that the step size values ​​fluctuate little around the mean, and the consistency and repeatability of the control process are high.

[0086] Simultaneously, the entire step size distribution lies entirely within the system noise fluctuation range of ±0.02 dB, as indicated by the red dashed line. This means that even the smallest control step size is significantly larger than the system's noise floor, ensuring that each programmed state is stable, reliable, and accurately distinguishable, effectively avoiding misjudgment or drift caused by noise interference. Furthermore, this centralized and controlled step size distribution is the result of the feedback mechanism described in step S30 of this application, which successfully compares the optical response change with a preset range and dynamically adjusts the pulse energy. This demonstrates that the feedback system can automatically correct and constrain the effects of most control events within an ideal, narrow range.

[0087] In summary, Figure 6 The statistical evidence provided fully demonstrates that the control method of this application can achieve high-precision and high-consistency multi-level state writing far exceeding the system noise level, which is a key characteristic for realizing high-capacity, high-precision neuromorphic optical computing hardware.

[0088] In this embodiment, after each optical pulse is applied, the system reads the transmittance in real time and dynamically calculates the amplitude and pulse width of the next pulse based on the deviation between the current value and the target value, achieving continuous and precise adaptive control of pulse-by-pulse. The effect of each pulse step is corrected in real time, ensuring the linearity of weight updates and the stability of multi-level states.

[0089] In this embodiment, when the transmittance change is small, the pulse width is reduced while the amplitude is increased; when the change is large, the pulse width is reduced and the amplitude is appropriately reduced. This strategy is based on the microscopic balance between the two competing physical processes of nucleation and growth in phase change materials. Shortening the pulse width can reduce heat accumulation and prevent overshoot, while increasing the amplitude ensures that the peak energy is sufficient to overcome the instantaneous nucleation barrier, thereby promoting uniform growth of crystal nuclei and enhancing the fine controllability of multi-level states.

[0090] The control system provided in this application is described below. The control system described below can be referred to in correspondence with the control method described above. In this embodiment, please refer to... Figure 7 The control system includes: The pulse generation module is used to generate an initial light pulse based on a preset initial energy parameter, and to determine the crystallization threshold light pulse that drives the phase change material layer to undergo a crystallization transition by adjusting the initial light pulse.

[0091] The execution and detection module is used to apply the crystallization threshold light pulse to the phase change material layer of the all-optical synaptic device and obtain the change in the optical response of the all-optical synaptic device.

[0092] The comparison and adjustment module is used to compare the optical response change with a preset change range, and when the optical response change is outside the preset change range, adjust the energy of the crystallization threshold light pulse based on PAWM.

[0093] The control module is used to control the execution and detection module and the comparison and adjustment module to execute cyclically until the optical state of the phase change material transitions to the target state through multiple intermediate states, thereby realizing multi-level state adjustment.

[0094] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the foregoing method embodiments, and will not be repeated here.

[0095] It should be understood that the above-mentioned control system is used to execute the methods in the above embodiments. The corresponding program modules in the control system are similar in implementation principle and technical effect to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0096] Based on the methods in the above embodiments, such as Figure 8 As shown in the illustration, this application provides a drive control device, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The processor can invoke logical instructions stored in the memory to execute the methods described in the above embodiments.

[0097] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0098] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0099] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0100] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0101] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0102] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0103] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0104] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for multi-level state control of an all-optical synaptic device based on PAWM modulation, characterized in that, include: Step S10: Based on the preset initial energy parameters, an initial light pulse is generated, and the crystallization threshold light pulse that drives the phase change material layer to undergo crystallization transformation is determined by adjusting the initial light pulse. Step S20: Apply the current crystallization threshold light pulse to the phase change material layer of the all-optical synapse device to obtain the change in the optical response of the all-optical synapse device; Step S30: Based on the comparison between the optical response change and a preset change range, and when the optical response change is outside the preset change range, adjust the energy of the crystallization threshold light pulse based on PAWM; Step S40: Repeat steps S20 and S30 until the optical state of the phase change material transitions to the target state through multiple intermediate states, thereby achieving multi-level state adjustment.

2. The method for multi-level state control of an all-optical synaptic device based on PAWM modulation as described in claim 1, characterized in that, Step S10 includes: An initial light pulse is generated based on a preset initial energy parameter, and a pulse sequence containing multiple light pulses with increasing power based on the initial light pulse is generated based on the initial light pulse. The pulse sequence is applied to the phase change material layer of the all-optical synaptic device, and the change in the optical response of the all-optical synaptic device after each applied light pulse is obtained; When the change in optical response changes, the current optical pulse is identified as the crystallization threshold optical pulse, and the application of subsequent optical pulses is stopped.

3. The multi-level state control method for an all-optical synaptic device based on PAWM modulation as described in claim 1, characterized in that, The preset range of change includes a first threshold and a second threshold, wherein the first threshold is less than the second threshold; step S30 includes: If the change in optical response is less than the first threshold, the parameters of the crystallization threshold light pulse are adjusted based on PAWM to increase the pulse energy; If the change in optical response is greater than the second threshold, the parameters of the crystallization threshold light pulse are adjusted based on PAWM to reduce the pulse energy.

4. The multi-level state control method for an all-optical synaptic device based on PAWM modulation as described in claim 3, characterized in that, The parameters of the crystallization threshold light pulse are adjusted based on PAWM to increase the pulse energy, specifically by decreasing the pulse width with a first preset gradient and increasing the pulse amplitude with a second preset gradient. The parameters of the crystallization threshold light pulse are adjusted based on PAWM to reduce the pulse energy. Specifically, the pulse width is increased by a first preset gradient and the pulse amplitude is decreased by a second preset gradient.

5. The multi-level state control method for an all-optical synaptic device based on PAWM modulation as described in any one of claims 1 to 4, characterized in that, The change in optical response is the change in transmittance; Specifically, the change in optical response is obtained as follows: Before and after applying an optical pulse to the phase change material layer, the output optical power of a probe beam after passing through the phase change material layer is measured respectively. The change in transmittance is determined based on the change in output optical power before and after the application of the optical pulse.

6. The method for multi-level state control of an all-optical synaptic device based on PAWM modulation as described in claim 1, characterized in that, Step S40 includes: Repeat steps S20 and S30 to obtain the intermediate state of the all-optical synaptic device in each cycle; When the optical response in the current intermediate state meets the preset target value, the current intermediate state is identified as the target state, and the loop stops.

7. A multi-level state control system for an all-optical synaptic device based on PAWM modulation, characterized in that, include: The pulse generation module is used to generate an initial light pulse based on a preset initial energy parameter, and to determine the crystallization threshold light pulse that drives the phase change material layer to undergo a crystallization transition by adjusting the initial light pulse. The execution and detection module is used to apply the crystallization threshold light pulse to the phase change material layer of the all-optical synaptic device and obtain the change in the optical response of the all-optical synaptic device. The comparison and adjustment module is used to compare the optical response change with a preset change range, and when the optical response change is outside the preset change range, adjust the energy of the crystallization threshold light pulse based on PAWM; The control module is used to control the execution and detection module and the comparison and adjustment module to execute cyclically until the optical state of the phase change material transitions to the target state through multiple intermediate states, thereby realizing multi-level state adjustment.

8. A multi-level state control device for an all-optical synaptic device based on PAWM modulation, characterized in that, Includes memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the control device to perform the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the control device, the control device performs the multi-level state control method for all-optical synaptic devices based on PAWM modulation as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are run on the drive control device, the drive control device performs the method as described in any one of claims 1 to 6.

Images