Mobile phone screen driving chip system

Abstract

The invention relates to the technical field of screen driving chips, and discloses a mobile phone screen driving chip system which comprises a parameter fusion module, a current field modeling module, a cooperative driving module, a dynamic compensation module and a self-consistency verification module. The method comprises the steps of generating a dynamic quantum efficiency field through fusion of a sensor and a deep neural network, solving a nonlinear current field equation to output a driving current gradient, constructing a weight matrix to optimize multi-screen cooperative data transmission, monitoring a brightness decay reverse correction boundary and a coupling coefficient in real time, and performing online iteration optimization parameter feedback modeling to realize a closed loop. Full-link closed-loop driving control is constructed through dynamic quantum efficiency field fusion, nonlinear discrete current field modeling, weight matrix compression transmission, time-varying attenuation factor reverse correction and self-consistent parameter optimization, and screen display uniformity enhancement, multi-screen cooperation efficiency improvement, aging effect real-time compensation and system long-term stability guarantee are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of screen driving chips, and specifically to a mobile phone screen driving chip system. Background Art

[0002] In the existing screen driving technology, ensuring display uniformity and optoelectronic response consistency highly depends on factory static calibration parameters, and it is difficult to adapt to the spatial differences in transistor mobility and the dynamic changes in quantum efficiency over time. Traditional current field modeling methods usually adopt linear or quasi-static assumptions, ignoring the complex non-linear coupling effects between pixels, resulting in insufficient prediction accuracy of the driving current distribution. Especially in dynamic picture and multi-screen collaboration scenarios, brightness jumps and color distortion are likely to occur.

[0003] In addition, fixed compression ratios and priority strategies are generally adopted for multi-screen data transmission, and it is impossible to dynamically optimize bandwidth allocation according to the picture content, resulting in high dynamic area refresh delays and detail loss. For compensating the screen aging effect, existing solutions mostly rely on periodic manual calibration or static adjustment based on empirical formulas, lacking the ability of real-time perception and closed-loop correction of local attenuation, and it is difficult to suppress the display non-uniformity caused by long-term use. At the same time, the optimization of current field control parameters is usually limited to offline calibration or manual debugging, and it is impossible to perform online iterative learning according to the actual driving error, resulting in insufficient model adaptability and limited long-term stability of the system.

[0004] Therefore, the present invention proposes a mobile phone screen driving chip system to solve the deficiencies of the existing technology. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a mobile phone screen driving chip system, which solves the problems of poor display uniformity, low multi-screen collaboration efficiency, and insufficient long-term stability caused by static parameter calibration, lack of non-linear coupling modeling, lagging aging compensation, and rigid parameter optimization.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A mobile phone screen driving chip system, the system includes:

[0007] A parameter fusion module, configured to respectively collect the transistor mobility distribution and quantum efficiency distribution of screen pixels through a transistor mobility sensor and an optoelectronic sensor built in the driving chip, and generate dynamic quantum efficiency field parameters through non-linear mapping fusion;

[0008] A current field modeling module, based on the dynamic quantum efficiency field parameters, solves a non-linear discrete current field equation including pixel coupling effects, and outputs the real-time driving current and its corresponding spatial gradient distribution;

[0009] The collaborative driving module dynamically constructs the weight matrix between screens according to the spatial gradient distribution, generates the multi-screen collaborative data stream through the sparse matrix compression algorithm, and assigns data transmission priorities based on the time-domain variation characteristics of the driving current;

[0010] The dynamic compensation module monitors the pixel brightness attenuation in real time through the optoelectronic sensor, combines the spatial distribution information of the weight matrix between screens, generates a time-varying attenuation factor, and reversely corrects the current field boundary conditions and pixel coupling coefficients of the current field modeling module;

[0011] The self-consistency verification module online iteratively optimizes the control parameters of the non-linear discrete current field equation according to the current field boundary conditions, coupling coefficients corrected by the dynamic compensation module, and data transmission errors of the collaborative driving module, and feeds the optimized parameters back to the current field modeling module.

[0012] Preferably, the parameter fusion module obtains the pixel-level transistor mobility distribution parameters through the transistor mobility sensor array built in the driving chip and calibrates the reference quantum efficiency distribution using the optoelectronic sensor during the screen initialization phase and establishes a non-linear mapping relationship of the dynamic quantum efficiency field parameters based on the historical driving current cumulative value which is specifically expressed as:

[0013] ;

[0014] where is the dynamic quantum efficiency field parameter of pixel ; is the non-linear mapping function implemented through deep neural network training, whose input is the transistor mobility and the historical driving current cumulative value and the output is the dynamic quantum efficiency value that combines the process deviation and the aging effect;

[0015] The historical driving current cumulative value is updated in real time through the current integrator of the driving chip, and the calculation formula is:

[0016] ;

[0017] where is the timestamp when the screen is first enabled; is the instantaneous driving current of pixel at time .

[0018] 3. Preferably, the current field modeling module calculates the driving current and its spatial gradient distribution in real time by solving the non-linear discrete current field equation, and the equation is defined as:

[0019] ;

[0020] Among them, represents the driving current of the pixel at time ; is the Laplacian operator of the current field, characterizing the spatial diffusion effect of the current on the screen plane; is a non - linear activation function related to the quantum efficiency, and its input is the current driving current and the dynamic quantum efficiency field parameter ; is the coupling coefficient between the pixel and the adjacent pixel and is used to describe the intensity of the current interaction between pixels; is a preset parameter for controlling the weights of the diffusion term, the non - linear term, and the coupling term; is the set of adjacent regions of the pixel .

[0021] Preferably, the calculation expression of the coupling coefficient is:

[0022] ;

[0023] Among them, and are the coordinates of the current pixel and the adjacent pixel respectively; is a very small constant to prevent the denominator from being zero, and its value range satisfies .

[0024] Preferably, the collaborative driving module dynamically constructs the inter - screen weight matrix based on the spatial gradient distribution output by the current field modeling module, and the matrix element is calculated as:

[0025] ;

[0026] Among them, represents the current spatial gradient of the pixel at time and is extracted from the gradient distribution data output by the current field modeling module; is a normalization parameter related to the physical size and resolution of the screen, used to adjust the weight decay rate; the matrix row and column indices correspond to the pixel block numbers of the transmitting - end and receiving - end screens respectively, and each block contains a plurality of continuously arranged pixels.

[0027] Preferably, the multi - screen collaborative data stream format generated by the sparse matrix compression algorithm is:

[0028] ;

[0029] Among them, is the luminance matrix of the current screen frame, and its element represents the target luminance value of the pixel ; is the coordinate offset matrix of adjacent screens, and its element represents the spatial position offset of the receiving - end screen pixel relative to the transmitting - end screen; is the Hadamard product operator, indicating element - by - element multiplication of matrix elements;

[0030] The data stream only transmits the and data blocks corresponding to non - zero elements, and compresses the coordinate indices of non - zero elements through run - length encoding.

[0031] Preferably, the dynamic compensation module real - time collects the actual luminance value of the pixel through a photoelectric sensor, and calculates the time - varying attenuation factor in combination with the theoretical luminance output by the current - field modeling module. Its update formula is:

[0032] ;

[0033] Among them, is the photoelectric conversion efficiency coefficient, determined by the factory - calibrated data of the screen; is the time - integration window length, used to smooth the luminance attenuation fluctuation; is the refresh period of the driving chip;

[0034] The initial value of the time - varying attenuation factor is set to 1, indicating a non - attenuation state.

[0035] Preferably, the reverse correction operation includes synchronous adjustment of the current - field boundary conditions and the pixel - to - pixel coupling coefficient. Specifically:

[0036] Boundary - condition correction: Inject the time - varying attenuation factor into the boundary conditions of the current - field modeling module. The correction formula is:

[0037] ;

[0038] Among them, is the spatial attenuation gradient compensation coefficient, used to suppress the luminance mutation caused by uneven aging at the screen edge;

[0039] Coupling - coefficient adjustment: According to the screen - to - screen weight matrix The spatial distribution information to dynamically scale the pixel - to - pixel coupling coefficient:

[0040] ;

[0041] Among them, is the weight matrix element of the transmitter block and the receiver block , which is used to strengthen or weaken the pixel coupling intensity in a specific area.

[0042] Preferably, the self - consistency verification module realizes the online iterative optimization of parameters through the following steps:

[0043] Error calculation: Real - time obtain the theoretical driving current of the current - field modeling module and the actual output current of the digital - to - analog converter (DAC) , and calculate the sum of the absolute values of the pixel - level current deviation:

[0044] ;

[0045] Gradient descent optimization: When exceeds the preset threshold , synchronously update the control parameters , , of the non - linear discrete current - field equation:

[0046] ;

[0047] ;

[0048] ;

[0049] Among them, is the adaptive learning rate, which is dynamically adjusted according to the variance of the historical error sequence:

[0050] ;

[0051] Among them, is the initial learning rate, is the historical error mean;

[0052] Parameter feedback: Transmit the optimized , , back to the current - field modeling module in real - time to replace the original control parameters.

[0053] The present invention also provides a method for driving a mobile phone screen chip, and the method includes the following steps:

[0054] S1. Collect the transistor mobility distribution and quantum efficiency distribution of the screen pixels through the built-in sensors of the driving chip. Combine the historical cumulative value of the driving current and use a deep neural network for non-linear mapping fusion to generate dynamic quantum efficiency field parameters;

[0055] S2. Based on the dynamic quantum efficiency field parameters, solve the non-linear discrete current field equation including the pixel coupling effect, and output the real-time driving current and its spatial gradient distribution;

[0056] S3. Construct an inter-screen weight matrix according to the spatial gradient distribution, generate a multi-screen collaborative data stream through a sparse matrix compression algorithm, and dynamically allocate data transmission priorities based on the temporal variation of the current;

[0057] S4. Monitor the pixel brightness attenuation in real time, generate a time-varying attenuation factor in combination with the spatial distribution of the weight matrix, and inversely correct the current field boundary conditions and the pixel coupling coefficient;

[0058] S5. According to the corrected boundary conditions, coupling coefficients and data transmission errors, online iteratively optimize the control parameters of the current field equation, and feedback the optimization results to the current field modeling process.

[0059] The present invention provides a mobile phone screen driving chip system. It has the following beneficial effects:

[0060] 1. Through the non-linear fusion technology of dynamic quantum efficiency field parameters, the present invention compensates for the pixel-level transistor mobility deviation and quantum efficiency attenuation in real time, and solves the problem of display non-uniformity caused by manufacturing process differences and long-term aging. Combining the multi-source data fusion ability of the deep neural network, it realizes high-precision optoelectronic response characteristic mapping, and significantly improves the screen color consistency and brightness uniformity.

[0061] 2. Based on the non-linear discrete current field equation constructed by the pixel coupling effect and combined with the hardware acceleration solution algorithm, the present invention can output the high-precision driving current distribution and spatial gradient data in real time. Through explicit iteration and parallel computing optimization, while ensuring the solution efficiency, this model accurately depicts the complex current diffusion and non-linear response characteristics, providing a reliable data basis for collaborative driving.

[0062] 3. Through the dynamic weight matrix and sparse matrix compression algorithm, the present invention significantly reduces the bandwidth requirements of the multi-screen collaborative data stream. Allocating transmission priorities based on the current temporal variation characteristics ensures low-latency refreshing of dynamic picture areas and efficient compression of static areas, realizing efficient resource scheduling and collaborative control of large-scale distributed screen arrays.

[0063] 4. Based on the reverse correction mechanism of time-varying attenuation factor and weight matrix, the present invention can suppress the distortion of current distribution at the screen edge and aging area in real time. By dynamically adjusting the boundary conditions and the coupling coefficient between pixels, an adaptive closed-loop compensation network is formed, effectively extending the screen service life and maintaining long-term display stability.

[0064] 5. Through the online iterative optimization and adaptive learning rate adjustment technology, the present invention realizes the dynamic self-correction of the current field control parameters. Combining the error feedback and gradient descent algorithms, it continuously improves the model prediction accuracy and system robustness, forming a full-link closed-loop control ability from data acquisition, modeling to optimization, ensuring the long-term accurate operation of the drive system. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is the system architecture diagram of the present invention;

[0066] Figure 2 is the method flow diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0068] Please refer to Figure 1 , the embodiment of the present invention provides a mobile phone screen driving chip system, and the system includes:

[0069] A parameter fusion module, configured to collect the transistor mobility distribution and quantum efficiency distribution of screen pixels respectively through a transistor mobility sensor and a photoelectric sensor built in the driving chip, and generate dynamic quantum efficiency field parameters through non-linear mapping fusion;

[0070] In this embodiment, the parameter fusion module generates dynamic quantum efficiency field parameters through multi-source heterogeneous sensor data acquisition, historical aging feature extraction, and non-linear deep learning model fusion. The following details the technical implementation details step by step:

[0071] The transistor mobility sensor built in the driving chip adopts a micro piezoresistive sensing unit, and each sensor corresponds to a thin film transistor (TFT) driving circuit of a pixel. By measuring the drain current and the gate voltage of the dynamic response, the mobility parameter is calculated, and the specific formula is:

[0072] ;

[0073] Among them, , are the length and width of the TFT channel respectively; is the capacitance per unit area of the gate oxide layer; is the differential value of the drain current with respect to the gate voltage, which is measured in real time by a differential amplifier circuit.

[0074] The sensor array refreshes data in a frame-synchronous manner, and the refresh rate is consistent with the screen driving frequency (typical value is 60 - 120 Hz), ensuring that the mobility distribution reflects the process deviation and stress aging effect of the pixel driving circuit in real time.

[0075] In the screen initialization stage, an optoelectronic sensor and an external high-precision spectrometer are used to calibrate the reference quantum efficiency collaboratively . The calibration process is as follows:

[0076] Constant current drive: Input a preset current to the pixel , and the duration is ;

[0077] Luminance measurement: The optoelectronic sensor collects the actual emitted luminance , and calculates the calibration value through the following formula:

[0078] ;

[0079] Among them, is the optoelectronic conversion efficiency coefficient preset at the factory and obtained through calibration with a standard light source; is the calibration current, and its value range is 20% - 80% of the maximum driving current of the screen.

[0080] Non-volatile storage: The calibration data is stored in the EEPROM of the driving chip for subsequent dynamic quantum efficiency calculation.

[0081] Each pixel is configured with an independent current integrator circuit for recording the cumulative value of the historical driving current . The integrator is implemented based on the principle of capacitor charge and discharge, and the specific calculation formula is:

[0082] ;

[0083] Among them, is the capacitance value of the integration capacitor; is the change in capacitor voltage; is the integration time window; is the timestamp when the screen is first enabled, recorded by the system clock module; is the pixel at time Instantaneous drive current

[0084] Integration result Reflects the cumulative aging degree of pixels in real time, and its value has a positive correlation with the TFT threshold voltage drift Showing a positive correlation and satisfying:

[0085] ;

[0086] Through a deep neural network model Fuse the mobility With the historical current To generate dynamic quantum efficiency field parameters The mathematical expression is:

[0087] ;

[0088] Wherein, Is the dynamic quantum efficiency field parameter of the pixel ; Is a non - linear mapping function realized through deep neural network training, whose input is the transistor mobility And the cumulative value of historical drive current , and the output is the dynamic quantum efficiency value that fuses process deviation and usage aging effect;

[0089] Details of the neural network structure:

[0090] Input layer: Dual - channel data, which are respectively the normalized And ;

[0091] Hidden layer: It contains 3 residual convolutional blocks, and each block consists of the following sub - layers: Convolutional layer (convolution kernel size , stride 1, padding 1); Batch normalization layer (Batch - Normalization); ReLU activation function;

[0092] Attention mechanism: Insert a channel attention module between the second and third residual blocks to dynamically adjust the feature weights through the SENet structure;

[0093] Output layer: The convolutional layer maps the features to a single channel, and the output range is constrained to [0, 1] by the Sigmoid function.

[0094] Training data and optimization:

[0095] Data source: Multi - batch screen aging experiment data, covering different environmental temperature (-20°C to 85°C) and humidity (10% - 90%RH) conditions;

[0096] Loss function: weighted combination of mean square error (MSE) and structural similarity (SSIM):

[0097] ;

[0098] wherein, , is the empirical weight coefficient;

[0099] Optimizer: AdamW optimizer is adopted, with an initial learning rate , and the training period is 300 epochs.

[0100] Data synchronization and real-time guarantee:

[0101] Hardware acceleration: Neural network inference is implemented through the NPU (Neural Network Processing Unit) built into the drive chip, supporting INT8 quantization operations, and the single inference latency is μs;

[0102] Frame synchronization mechanism: The dynamic quantum efficiency field parameter is updated during the vertical blanking interval of each frame, and is strictly synchronized with the screen refresh rate; Exception handling: When sensor data is abnormal (such as mobility exceeding the limit or current integration overflow), backup parameters are enabled and the system self-check is triggered.

[0103] The current field modeling module, based on the dynamic quantum efficiency field parameter, solves the non-linear discrete current field equation including the pixel coupling effect, and outputs the real-time drive current and its corresponding spatial gradient distribution;

[0104] In this embodiment, the current field modeling module calculates the real-time drive current distribution and its spatial gradient at the screen pixel level by establishing a non-linear discrete current field equation and adopting an efficient numerical solution method. The following discloses the technical details in full from the aspects of equation construction, parameter definition, solution algorithm and hardware implementation:

[0105] The core control equation of the current field modeling module is a spatio-temporal discretized partial differential equation, and its complete form is defined as:

[0106] ;

[0107] wherein, represents the drive current of pixel at time ; is the Laplacian operator of the current field, characterizing the spatial diffusion effect of the current on the screen plane; is a non-linear activation function related to quantum efficiency, and its input is the current drive current And dynamic quantum efficiency field parameters ; For a pixel And adjacent pixels The coupling coefficient between them, which is used to describe the strength of the current interaction between pixels. The calculation formula is: , where And Are the coordinates of the current pixel and the adjacent pixel respectively, Is a very small constant to prevent the denominator from being zero, and its value range satisfies ; Is a preset parameter for controlling the weights of the diffusion term, the nonlinear term and the coupling term; For a pixel The set of adjacent regions;

[0108] Numerical solution algorithm and stability guarantee:

[0109] Time discretization method: The explicit Euler method is used to discretize the partial differential equation in time. The iteration formula is:

[0110] ;

[0111] Among them, the superscript And Respectively represent the current values at the th and the th time steps; Is the time step (unit: s), and its value is restricted by the CFL (Courant - Friedrichs - Lewy) stability condition:

[0112] ; Ensure numerical convergence and no oscillation.

[0113] Spatial gradient calculation: While solving the current distribution , calculate its spatial gradient , which is used to construct the weight matrix of the co - driving module. The gradient calculation uses the Sobel operator discretization:

[0114] ;

[0115] ;

[0116] ;

[0117] Boundary condition handling: Set the normal gradient of the current on the four sides of the screen to zero (Neumann boundary condition), that is:

[0118] ;

[0119] This condition is achieved by the mirror extension method, that is, when calculating the boundary pixels, the current value of the out-of-bounds index is replaced with the value of the adjacent internal pixel. For example:

[0120] ;

[0121] Among them, is the total number of pixels in the horizontal direction of the screen.

[0122] Hardware acceleration and real-time implementation:

[0123] Parallel computing architecture: The driver chip integrates a dedicated parallel computing unit, adopting the SIMD (Single Instruction Multiple Data) architecture. Each computing core is responsible for the current field calculation task of a pixel block (typically 8×8 pixels). The following optimizations are implemented inside the computing unit:

[0124] Data locality optimization: The current values and coupling coefficients of adjacent pixels are stored in the on-chip high-speed cache to reduce the access latency of external memory;

[0125] Pipeline design: The calculations of the diffusion term, nonlinear term, and coupling term are executed in stages in a pipeline to improve the instruction throughput rate.

[0126] Exception handling mechanism: Overcurrent protection: When it is detected that ( is the maximum allowable current of the drive circuit), the output current is immediately clamped to , and an interrupt service program is triggered to update the control parameters ;

[0127] Numerical overflow detection: A saturating arithmetic unit is used to handle floating-point operation overflows to ensure calculation stability.

[0128] Adaptive adjustment of control parameters: The control parameter is dynamically optimized in the following way:

[0129] Initial calibration: When the screen leaves the factory, based on the current response data under the drive of standard test patterns (such as all white, all black, checkerboard), the least squares method is used to fit the optimal parameter combination;

[0130] Online update: The self-consistency verification module adjusts the parameters in real time according to the DAC output error, and the update strategy is:

[0131] ;

[0132] Among them, the partial derivative is calculated in real time through the automatic differentiation (Auto-Diff) technology.

[0133] The collaborative driving module dynamically constructs an inter-screen weight matrix according to the spatial gradient distribution, generates a multi-screen collaborative data stream through a sparse matrix compression algorithm, and assigns data transmission priorities based on the time-domain change characteristics of the driving current;

[0134] In this embodiment, the collaborative driving module dynamically constructs an inter-screen weight matrix based on the spatial gradient distribution data output by the current field modeling module, generates a multi-screen collaborative data stream through a sparse compression algorithm, and assigns data transmission priorities in combination with the time-domain change characteristics of the current. The following fully discloses the technical details from the aspects of weight matrix construction, data stream compression, and transmission optimization:

[0135] Dynamically constructing the inter-screen weight matrix, the weight matrix whose matrix elements are determined by the current gradient correlation between the transmitting-end pixel block and the receiving-end pixel block , and its calculation formula is:

[0136] ;

[0137] wherein, is the current spatial gradient amplitude (unit: A / m) of the reference pixel of the transmitting-end pixel block at time , which is directly extracted from the gradient distribution data output by the current field modeling module; is the normalized attenuation coefficient (unit: dimensionless), and its value satisfies the relationship with the physical size of the screen (unit: m) and the resolution (unit: pixels per inch), where is a proportional constant, preferably determined by experimental calibration; : matrix row and column indices, corresponding to the pixel block numbers of the transmitting-end and receiving-end screens respectively, and each block contains continuously arranged pixels (typical ), and the block division method is consistent with the physical partition of the screen; is the Euclidean norm operator, used to quantify the spatial change intensity of the current gradient.

[0138] Physical meaning of the weight matrix:

[0139] The weight value corresponding to the high-gradient region (such as the edge of the dynamic picture) is relatively low, reducing the data transmission volume in such regions;

[0140] The weight value corresponding to the low-gradient region (such as the static background) is relatively high, retaining more data details to ensure display smoothness.

[0141] The sparse matrix compression algorithm of the multi-screen collaboration data stream compression algorithm combines the weight matrix with the screen brightness matrix , the coordinate offset matrix to generate a compressed data stream , and its mathematical expression is:

[0142] ;

[0143] Among them, is the brightness matrix of the current screen frame, and the element represents the target brightness value (unit: nit) of the pixel , which is output by the display content rendering engine; is the coordinate offset matrix of the adjacent screen, and the element represents the spatial position offset (unit: pixel) of the receiving end screen pixel relative to the reference point of the sending end screen, and the offset calculation method is:

[0144] ;

[0145] Among them, is the reference coordinate of the sending end screen in the global coordinate system; is the Hadamard-Product operator, indicating element-by-element multiplication of the matrix; is the standard matrix multiplication operator.

[0146] Data compression and transmission rules:

[0147] Non-zero element screening: Only transmit the non-zero elements in (that is, ), and the threshold is dynamically adjusted according to the bandwidth limit) corresponding and data blocks;

[0148] Coordinate index encoding: Use run-length encoding (RLE) to compress the block coordinates of non-zero elements. The encoding rule is: Traverse the weight matrix along the screen scan line direction (from left to right, from top to bottom); Record the start position and length of the continuous zero element interval, and directly record the coordinates and data of non-zero elements;

[0149] Data packet encapsulation: The compressed data stream is organized according to the frame structure, including a frame header (timestamp, screen ID), a payload (compressed data), and a checksum (CRC32).

[0150] Dynamic allocation of data transmission priority: Based on the time-domain variation characteristics of the driving current, different transmission priorities are assigned to different pixel blocks. The specific implementation method is as follows:

[0151] Calculation of time-domain change rate: For each pixel block , calculate the change rate of its driving current within the time window :

[0152] ;

[0153] Among them, is the pixel set within the block ;

[0154] Priority mapping: Map the time-domain change rate to the transmission priority level , and the mapping relationship is:

[0155] ;

[0156] Among them, is the total number of priority levels (typical ); and are respectively the minimum and maximum time-domain change rates of all blocks in the current frame;

[0157] Transmission scheduling: High-priority data is directly transmitted through a dedicated physical layer channel, and low-priority data is transmitted using idle bandwidth to ensure the dynamic area refresh rate and display real-time performance.

[0158] Hardware implementation and optimization:

[0159] Parallel processing unit: Preferably, the construction of the weight matrix and data compression are implemented through the built-in parallel computing unit of the driving chip, supporting multi-threaded concurrent processing;

[0160] Dedicated encoder: Run-length encoding and packet encapsulation are implemented using a hardware accelerator, and the encoding delay is less than 1 microsecond;

[0161] Dynamic bandwidth allocation: The transmission priority scheduler adjusts and in real time according to the network congestion status to balance data quality and transmission delay.

[0162] The dynamic compensation module monitors the pixel brightness attenuation in real time through the photoelectric sensor, combines the spatial distribution information of the weight matrix between the screens, generates a time-varying attenuation factor, and reversely corrects the current field boundary conditions and pixel-to-pixel coupling coefficients of the current field modeling module;

[0163] In this embodiment, the dynamic compensation module monitors the pixel brightness attenuation in real time through a photoelectric sensor, combines the spatial distribution information of the weight matrix between screens, generates a time-varying attenuation factor, and reversely corrects the boundary conditions and pixel-to-pixel coupling coefficients of the current field modeling module, thereby suppressing the display non-uniformity caused by screen aging. The following discloses the technical details in full from the aspects of brightness monitoring, attenuation factor calculation, boundary condition correction, and coupling coefficient adjustment:

[0164] Real-time brightness monitoring and attenuation factor calculation: The dynamic compensation module collects the actual pixel brightness values through the photoelectric sensor array built into the driving chip and compares them with the theoretical brightness values output by the current field modeling module, and dynamically updates the time-varying attenuation factor . The specific implementation method is as follows:

[0165] Theoretical brightness calculation: According to the driving current output by the current field modeling module and the dynamic quantum efficiency field parameters generated by the parameter fusion module, calculate the theoretical brightness:

[0166] ;

[0167] Among them, is the photoelectric conversion efficiency coefficient (unit: nit / A), which is determined by the factory calibration data of the screen. The calibration method is to input a known current under standard lighting conditions and measure the actual brightness and then calculate , is the factory reference quantum efficiency; is the dynamic quantum efficiency parameter of pixel , which is generated in real time by the parameter fusion module; is the driving current of pixel at time (unit: A). Update of the time-varying attenuation factor: Based on the deviation between the actual brightness and the theoretical brightness, the sliding time window integration algorithm is used to update the attenuation factor:

[0168] ;

[0169] Among them, is the refresh period of the driving chip (unit: s), which is strictly synchronized with the screen refresh rate. For example, for a 120Hz refresh rate, ; is the length of the time integration window (unit: s), which is used to smooth the instantaneous brightness fluctuations, and its value range is ( is the number of refresh frames included in the window, and the typical value is 10 - 100); It is the initial condition, indicating the non - attenuation state.

[0170] Sensor data calibration: The brightness data collected by the optoelectronic sensor needs to be compensated for dark current and corrected for non - linearity. The dark current compensation formula is:

[0171] ;

[0172] where is the background brightness without drive current, which is measured and updated through a periodic self - test process.

[0173] Inverse correction of the current field boundary condition: Inject the time - varying attenuation factor into the boundary condition of the current field modeling module to suppress the brightness mutation caused by uneven aging at the screen edge. The corrected boundary current is calculated as:

[0174] ;

[0175] where is the spatial attenuation gradient compensation coefficient (unit: m·A / nit), whose physical meaning is the contribution weight of the spatial gradient of the attenuation factor to the boundary current, and is determined by fitting the aging experimental data; is the spatial gradient of the attenuation factor in the horizontal direction of the screen, and the discretized calculation uses the central difference method:

[0176] ;

[0177] where is the pixel pitch (unit: m), which is calculated from the physical size and resolution of the screen.

[0178] Boundary condition update mechanism: The corrected boundary current will replace the boundary value of the original current field modeling module, which is specifically implemented through the following steps:

[0179] Boundary pixel identification: Determine the left / right / top / bottom boundary pixel sets according to the physical coordinates of the screen;

[0180] Gradient calculation: For each boundary pixel, calculate the spatial gradient of its attenuation factor;

[0181] Current correction: Update the boundary current value according to the formula and write it into the input buffer of the current field modeling module.

[0182] Dynamic adjustment of the pixel - to - pixel coupling coefficient: Combining the spatial distribution information of the screen - to - screen weight matrix dynamically scale the pixel - to - pixel coupling coefficient to adapt to the change in coupling strength caused by aging. The adjustment formula is:

[0183] ;

[0184] Among them, is the weight matrix element of the sender block and the receiver block , which is generated by the collaborative driving module, and the calculation method is:

[0185] ;

[0186] The division rule of block and block is: divide the screen into blocks of pixels (typical ), and the block numbers are arranged in row-major order;

[0187] The calculation method of the original coupling coefficient is:

[0188] ;

[0189] Among them, is a very small constant (unit: dimensionless) to prevent the denominator from being zero, and its value range satisfies .

[0190] Coupling coefficient update process:

[0191] Weight matrix matching: According to the pixels and the blocks and block to which they belong, query the weight matrix element ;

[0192] Coefficient scaling: Multiply the original coupling coefficient by to obtain the adjusted coefficient ;

[0193] Data synchronization: The updated coupling coefficient is written into the coefficient register of the current field modeling module through shared memory or a bus.

[0194] Data synchronization and real-time guarantee:

[0195] Sensor data synchronization: The optoelectronic sensor array collects luminance data in frames and synchronizes with the arithmetic unit of the dynamic compensation module through the on-chip bus. The data synchronization mechanism includes:

[0196] Frame start signal triggering: At the beginning of the screen vertical blanking period (VBlank), trigger the batch transmission of sensor data;

[0197] Dual-buffer design: Adopt a strategy of separating the front-end and back-end caches. When the front-end cache receives new data, the back-end cache is used for the computing unit to read, avoiding data conflicts.

[0198] Gradient calculation optimization: Spatial gradient The calculation is implemented through a hardware accelerator, specifically using a pre-stored difference kernel for convolution operation:

[0199] ;

[0200] Among them, is the horizontal gradient operator, and when implemented in hardware, a pipeline architecture is used to calculate the gradient values of all pixels in parallel.

[0201] Exception handling mechanism: Attenuation factor overrun detection: When or , it is determined as an outlier;

[0202] Historical data replacement: Enable the mean value of the attenuation factors of the most recent frames (typical ) to replace the outliers:

[0203] ;

[0204] System self-check: When the outliers continuously appear more than the threshold number of times, trigger the sensor calibration and circuit self-check procedures.

[0205] A self-consistency verification module, according to the current field boundary conditions, coupling coefficients, and data transmission errors of the collaborative driving module corrected by the dynamic compensation module, iteratively optimizes the control parameters of the non-linear discrete current field equation online, and feeds the optimized parameters back to the current field modeling module;

[0206] In this embodiment, the self-consistency verification module iteratively optimizes the control parameters of the non-linear discrete current field equation online by analyzing the deviation between the theoretical driving current and the actual output current of the current field modeling module in real time, and feeds the optimization result back to the modeling module to achieve dynamic parameter correction. The following discloses the technical details in full from the aspects of error calculation, gradient optimization, learning rate adjustment, and parameter feedback:

[0207] Error calculation and deviation analysis: The module obtains the theoretical driving current output by the current field modeling module and the actual output current of the digital-to-analog converter (DAC) in real time, and calculates the sum of the absolute values of the pixel-level current deviations as the optimization objective function:

[0208] ;

[0209] Among them, It is the theoretical driving current value (unit: A) calculated by the current field modeling module based on the boundary conditions and coupling coefficients corrected by the dynamic compensation module. Its calculation process includes numerical solutions of the diffusion term, non-linear term, and coupling term; It is the actual driving current value output by the digital-to-analog converter (unit: A), which is measured in real time through the current sampling circuit of the driving chip, and the sampling frequency is synchronized with the screen refresh rate; It is time The total global current deviation (unit: A), which is used to quantify the model prediction accuracy. When it triggers the parameter optimization process; : The preset error threshold (unit: A), which is dynamically adjusted according to 1% - 5% of the maximum driving current of the screen.

[0210] Data synchronization mechanism:

[0211] The theoretical current and actual current data are synchronously transmitted to the self-consistency verification module through the on-chip high-speed serial bus (such as AXI-Stream), and the transmission delay is less than 1% of the refresh period. Preferably, a double-buffer storage strategy is adopted. When the front-end buffer receives new data, the back-end buffer is used for the computing unit to read, avoiding data competition. The double-buffer switching signal is triggered by the vertical blanking period (VBlank) to ensure strict timing alignment.

[0212] When the error exceeds the preset threshold it triggers the gradient descent algorithm to synchronously update the control parameters of the non-linear discrete current field equation. The parameter update formula is:

[0213] ;

[0214] ;

[0215] ;

[0216] Among them, , , are the control parameters of the diffusion term, non-linear term, and coupling term (unit: dimensionless) for the th iteration, and the initial values are determined by calibrating the electrical characteristics of the screen at the factory; , , are the partial derivatives of the error with respect to the control parameters (unit: A -1 ), which are calculated in real time through the automatic differentiation (AutoDiff) technology. Specifically, the reverse-mode differential chain rule is adopted, and the calculation process is as follows:

[0217] Forward propagation calculation For Dependency relationship;

[0218] Backpropagation calculates the error For Gradient of ; Obtained by chain rule differentiation ;

[0219] is the adaptive learning rate (unit: dimensionless), dynamically adjusted to balance the convergence speed and stability.

[0220] Adaptive learning rate adjustment strategy: The learning rate is dynamically adjusted according to the statistical characteristics of the historical error sequence, and the calculation formula is:

[0221] ;

[0222] Among them, is the initial learning rate (unit: dimensionless), determined through off-line calibration experiments, and the typical value range is ; is the error value (unit: A) of the th iteration, stored in a circular buffer, and the buffer length is dynamically set according to the screen refresh rate (typical ); is the historical error mean (unit: A), and the calculation method is ;

[0223] Denominator term: The variance estimation term of the historical error, used to suppress the parameter oscillation caused by noise. When the error variance increases significantly, the learning rate is automatically reduced to enhance the algorithm stability.

[0224] Numerical stability guarantee: To avoid the learning rate explosion caused by too small denominator, a very small constant is added to correct the denominator:

[0225] ;

[0226] Parameter feedback and real-time update: The optimized control parameters , , are fed back to the current field modeling module through the following mechanism:

[0227] Parameter register writing: Write the new parameters into the configuration register of the current field modeling module, overwriting the original parameter values. The register address is mapped to memory-mapped I / O (MMIO) and updated directly through the write operation instruction;

[0228] Timing Synchronization: The parameter update operation is strictly limited to be completed during the vertical blanking period (VBlank) of the screen to avoid data conflicts during the refresh process. The vertical blanking signal is generated by the timing controller (TCON);

[0229] Rollback Mechanism: If the error continues for consecutive (typical

[0230] iterations without decreasing, it is determined that the optimization has failed, and it automatically reverts to the previous valid parameter group and triggers a system diagnostic interrupt;

[0231] Version Management: Each parameter update records the version number and timestamp and stores them in a non-volatile memory (such as Flash), supporting rollback by version during fault recovery. 、 、 Exceeds the threshold when, the parameter update is paused and backup parameters are enabled;

[0232] Error Divergence Protection: If the error does not decrease after consecutive iterations, it is determined that the optimization has failed, and the control parameters are reset to the factory calibration values. The factory values are stored in a read-only memory (ROM) and are protected from tampering by hardware fuses;

[0233] Logging: All parameter update records (including timestamps, error values, learning rates, partial derivatives) are stored in a non-volatile memory in binary format, and the storage interval is configurable (such as per frame or every 10 frames) for subsequent fault analysis.

[0234] Hardware Acceleration and Parallel Computing:

[0235] Automatic Differentiation Accelerator: A dedicated hardware unit is used to implement partial derivative calculations, supporting parallel processing of multi-pixel gradient chain differentiation. The internal of the calculation unit adopts a pipeline design to complete the gradient calculation of a single pixel in a single cycle;

[0236] Variance Calculation Optimization: The historical error variance calculation is implemented through a sliding window accumulator. When each new error value enters the buffer, the cumulative sum and are automatically updated to avoid repeated traversal of data;

[0237] Parameter Update Pipeline: The learning rate calculation, gradient descent update, and parameter write-back operations are executed in a three-stage pipeline, with each stage of the pipeline corresponding to one clock cycle, ensuring that the single iteration delay is less than 3 clock cycles.

[0238] Please refer to Figure 2, the present invention also provides a method for a mobile phone screen driving chip, and the method includes the following steps:

[0239] S1. Through the transistor mobility sensor array built in the driving chip, the mobility distribution data of each pixel is collected in real time, and the initial quantum efficiency distribution is calibrated in cooperation with an external spectrometer by using a photoelectric sensor. Combining the cumulative aging characteristics recorded by the historical driving current integrator, the multi-source heterogeneous data is input into a deep neural network model for non-linear fusion. The neural network adopts a multi-layer residual convolution structure to extract the correlation characteristics between mobility deviation, quantum efficiency decay, and current aging, and outputs dynamic quantum efficiency field parameters to reflect the dynamic changes of pixel-level optoelectronic response characteristics in real time.

[0240] S2. Based on the dynamic quantum efficiency field parameters, a discretized partial differential equation including current diffusion between pixels, non-linear response of quantum efficiency, and coupling effect of adjacent pixels is constructed. An explicit numerical iteration algorithm is used to solve the equation to calculate the driving current value of each pixel in real time and synchronously output the spatial gradient distribution of the current in the screen plane. During the solving process, a hardware acceleration unit is used to achieve efficient parallel calculation of the diffusion term, non-linear term, and coupling term, ensuring that the single-frame solving time is strictly less than the screen refresh period.

[0241] S3. According to the current spatial gradient distribution, a screen-to-screen weight matrix is dynamically constructed to quantify the priority of data transmission in different regions. Low weights are assigned to high-gradient regions (such as the edges of dynamic images) to reduce the amount of data, and high weights are assigned to low-gradient regions (such as static backgrounds) to retain details. The data blocks corresponding to non-zero weights are screened by a sparse matrix compression algorithm, and combined with run-length encoding to compress the coordinate index to generate a compact multi-screen collaborative data stream. At the same time, the transmission channel resources are dynamically allocated based on the time-domain change rate of the driving current to ensure the real-time performance and low latency of data in high-dynamic regions.

[0242] S4. The actual brightness decay of pixels is monitored in real time through the photoelectric sensor array, and the deviation between the theoretical brightness and the measured value is compared to generate a time-varying decay factor. This factor combines the spatial distribution information of the weight matrix to reversely correct the boundary conditions of the current field modeling module, suppress the current distortion caused by aging at the screen edge, and dynamically adjust the coupling coefficient between pixels to balance the current interaction strength between the aging region and the non-aging region. The corrected parameters are synchronously transmitted to the current field solver in real time through the bus to form a closed-loop compensation mechanism.

[0243] S5. The actual output current of the real-time acquisition digital-to-analog converter is collected and compared pixel by pixel with the theoretical value of the current field modeling module, and the total global deviation is calculated as the optimization target. When the deviation exceeds the preset threshold, the adaptive gradient descent algorithm is used to iteratively update the control parameters of the diffusion term, the non-linear term, and the coupling term. During the optimization process, the learning rate is dynamically adjusted according to the historical error variance to balance the convergence speed and stability. Finally, the optimized parameters are fed back to the modeling module to achieve continuous self-correction of the system parameters and long-term stability guarantee.

[0244] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A mobile phone screen driver chip system, characterized in that, The system includes: A parameter fusion module, which is used to collect the transistor mobility distribution and quantum efficiency distribution of screen pixels respectively through the transistor mobility sensor and photoelectric sensor built in the driving chip, and generate dynamic quantum efficiency field parameters through non-linear mapping fusion; A current field modeling module, which solves the non-linear discrete current field equation including the coupling effect between pixels based on the dynamic quantum efficiency field parameters, and outputs the real-time driving current and its corresponding spatial gradient distribution; A cooperative driving module, which dynamically constructs a weight matrix between screens according to the spatial gradient distribution, generates a multi-screen cooperative data stream through a sparse matrix compression algorithm, and assigns data transmission priorities based on the time-domain change characteristics of the driving current; A dynamic compensation module, which monitors the pixel brightness attenuation in real time through the photoelectric sensor, combines the spatial distribution information of the weight matrix between screens, generates a time-varying attenuation factor, and reversely corrects the current field boundary conditions and the coupling coefficient between pixels of the current field modeling module; A self-consistency verification module, which online iteratively optimizes the control parameters of the non-linear discrete current field equation according to the current field boundary conditions, coupling coefficients corrected by the dynamic compensation module and the data transmission error of the cooperative driving module, and feeds back the optimized parameters to the current field modeling module.

2. The mobile phone screen driving chip system according to claim 1, characterized in that The parameter fusion module obtains pixel-level transistor mobility distribution parameters through the transistor mobility sensor array built in the driving chip , and at the same time uses a photoelectric sensor to calibrate the reference quantum efficiency distribution during the screen initialization stage , and based on the historical driving current cumulative value Establish a non-linear mapping relationship of dynamic quantum efficiency field parameters, which is specifically expressed as: ； Among them, is the dynamic quantum efficiency field parameter of the pixel ; is a non-linear mapping function implemented through deep neural network training, with its input being the transistor mobility and the historical drive current cumulative value , and the output being the dynamic quantum efficiency value that fuses process deviation and uses the aging effect; The historical drive current cumulative value is updated in real time by the current integrator of the drive chip, and the calculation formula is: ； Among them, is the timestamp when the screen is first enabled; is the pixel at time instantaneous drive current.

3. The mobile phone screen driving chip system according to claim 1, characterized in that The current field modeling module calculates the driving current and its spatial gradient distribution in real time by solving the non-linear discrete current field equation, and the equation is defined as: ； Among them, represents the pixel driving current at time ; is the Laplacian operator of the current field, characterizing the spatial diffusion effect of the current in the screen plane; is a non - linear activation function related to the quantum efficiency, and its input is the current driving current and the dynamic quantum efficiency field parameter ; is the coupling coefficient between the pixel and the adjacent pixel , used to describe the intensity of the current interaction between pixels; is a preset parameter for controlling the weights of the diffusion term, non - linear term, and coupling term; is the set of adjacent regions of the pixel .

4. The mobile phone screen driving chip system according to claim 3, characterized in that, The coupling coefficient is calculated by the following expression: ； wherein, and are the coordinates of the current pixel and the adjacent pixel, respectively; is a very small constant to prevent the denominator from being zero, and its value range satisfies .

5. The mobile phone screen driving chip system according to claim 1, characterized in that The collaborative driving module dynamically constructs an inter-screen weight matrix based on the spatial gradient distribution output by the current field modeling module , and its matrix elements are calculated as follows: ； Among them, represents the pixel current spatial gradient at time , which is extracted from the gradient distribution data output by the current field modeling module; is a normalization parameter related to the physical size and resolution of the screen, and is used to adjust the weight decay rate; the matrix row and column indices respectively correspond to the pixel block numbers of the sender and receiver screens, and each block contains multiple pixels arranged continuously.

6. The mobile phone screen driving chip system according to claim 5, characterized in that, The format of the multi-screen cooperative data stream generated by the sparse matrix compression algorithm is: ； Among them, is the brightness matrix of the current screen frame, and its elements represent the target brightness value of the pixel ; is the coordinate offset matrix of the adjacent screen, and its elements represent the spatial position offset of the receiving - end screen pixel relative to the sending - end screen; is the Hadamard product operator, indicating element - by - element multiplication of matrix elements. The data stream only transmits corresponding to the non-zero elements in and data blocks, and compresses the coordinate indices of non-zero elements by run-length encoding.

7. A mobile phone screen driving chip system according to claim 1, characterized in that The dynamic compensation module real-time collects the actual brightness value of pixels through a photoelectric sensor , and combines with the theoretical brightness output by the current field modeling module to calculate the time-varying attenuation factor , and its update formula is: ； Among them, is the optoelectronic conversion efficiency coefficient, which is determined by the calibration data when the screen leaves the factory; is the time integration window length, which is used to smooth the brightness decay fluctuation; is the refresh period of the driving chip; The time-varying attenuation factor is initially set to 1, indicating a non-attenuated state.

8. A mobile phone screen driving chip system according to claim 1, characterized in that, The reverse correction operation includes synchronous adjustment of the current field boundary conditions and the coupling coefficient between pixels, specifically: Boundary condition correction: Inject the time-varying attenuation factor into the boundary conditions of the current field modeling module, and the correction formula is: ； Among them, is the spatial attenuation gradient compensation coefficient, which is used to suppress the brightness mutation caused by uneven aging at the screen edge; Coupling coefficient adjustment: Based on the spatial distribution information of the weight matrix between screens dynamically scale the coupling coefficient between pixels: ； Among them, is the weight matrix element of the sending-end block and the receiving-end block for strengthening or weakening the pixel coupling intensity in a specific region.

9. A mobile phone screen driving chip system according to claim 1, characterized in that The self-consistency verification module realizes online iterative optimization of parameters through the following steps: Error calculation: Obtain the theoretical driving current of the current field modeling module in real time and the actual output current of the digital-to-analog converter (DAC) , and calculate the sum of the absolute values of the pixel-level current deviations: ； Gradient descent optimization: When exceeds the preset threshold the control parameters of the non-linear discrete current field equation , , are synchronously updated: ； ； ； wherein, is the adaptive learning rate, which is dynamically adjusted according to the variance of the historical error sequence: ； Among them, is the initial learning rate, is the mean value of historical errors; Parameter feedback: The optimized , , are transmitted back to the current field modeling module in real time to replace the original control parameters.

10. A method for a mobile phone screen driving chip, applied to the system according to any one of claims 1-9, characterized in that, The method includes the following steps: S1. Collect the transistor mobility distribution and quantum efficiency distribution of screen pixels through the sensors built in the driving chip, combine the historical cumulative value of the driving current, and use a deep neural network for non-linear mapping fusion to generate dynamic quantum efficiency field parameters; S2. Based on the dynamic quantum efficiency field parameters, solve the non-linear discrete current field equation including the coupling effect between pixels, and output the real-time driving current and its spatial gradient distribution; S3. Construct a weight matrix between screens according to the spatial gradient distribution, generate a multi-screen cooperative data stream through a sparse matrix compression algorithm, and dynamically assign data transmission priorities based on the time-domain change of the current; S4. Monitor the pixel brightness attenuation in real time, generate a time-varying attenuation factor by combining the spatial distribution of the weight matrix, and reversely correct the current field boundary conditions and the coupling coefficient between pixels; S5. Online iteratively optimize the control parameters of the current field equation according to the corrected boundary conditions, coupling coefficients and data transmission error, and feed back the optimization results to the current field modeling process.

Images