Intelligent power management integrated circuit with AI dynamic adjustment and multi-mode cooperation and control method thereof

Abstract

The invention relates to an intelligent power management integrated circuit with AI dynamic adjustment and multi-mode cooperation and a control method thereof, and belongs to the field of power management integrated circuits, the intelligent power management integrated circuit comprises an AI dynamic adjustment module, a multi-protocol adaptive charging system and a three-dimensional heterogeneous packaging structure, the AI dynamic adjustment module comprises a hybrid neural network architecture and a federated learning framework, the multi-protocol self-adaptive charging system integrates a PD 3.1 EPR wired protocol control unit, a Qi v2.0 wireless charging driving module and a dynamic impedance matching circuit, and the three-dimensional heterogeneous packaging structure comprises a TVS (Transient Voltage Suppressor), a micro-channel heat dissipation unit and a FeRAM (FeRAM) storage unit. According to the invention, an AI dynamic adjusting system, a multi-protocol adaptive charging system, a three-dimensional heterogeneous packaging framework and a self-powered safety system are added, so that a novel AI intelligent power management chip which is low in power consumption, intelligent and small in size is truly realized.

Description

Technical Field

[0001] The present invention relates to an intelligent power management integrated circuit with AI dynamic regulation and multimodal collaboration and a control method thereof, and belongs to the field of power management integrated circuits. Background Art

[0002] Power management integrated circuits (PMICs) are highly integrated circuits used to optimize power distribution and management, thereby improving the energy efficiency and stability of electronic devices. With the rapid development of mobile devices, IoT devices, electric vehicles, and other fields, intelligent power management is becoming increasingly important in improving product performance, extending battery life, and reducing power consumption.

[0003] Existing PMICs have the following technical bottlenecks: High static power consumption: Traditional linear voltage regulation solutions are inefficient, with significant energy loss especially in low-load scenarios.

[0004] Insufficient dynamic response: The power supply strategy cannot be adjusted in real time according to the device's operating status (such as gaming or standby), resulting in limited battery life.

[0005] Single function: Most PMICs only support a single charging protocol (such as QC or PD), making it difficult to adapt to multi-modal charging requirements.

[0006] Safety risks: Lack of intelligent protection mechanisms for over-temperature, over-voltage, and short circuits, which can easily cause equipment damage.

[0007] Insufficient energy efficiency: The efficiency of traditional linear regulators is less than 40% at light loads, and the dynamic response delay of Buck-Boost converters is >100μs.

[0008] Protocol fragmentation: A single PMIC cannot simultaneously support USB PD 3.1 EPR (140W), Qi v2.0 (15W), and wireless reverse charging.

[0009] Thermal management defects: Package heat dissipation relies on an external heat sink, and the chip junction temperature fluctuates by >±15°C, affecting the lifespan.

[0010] Lack of intelligence: Static power supply strategies cannot predict sudden loads (such as 5G data bursts), resulting in voltage drops of >5%.

[0011] Existing power management integrated circuits still have some flaws. For example, in the invention patent with publication number US20220109345A1, the AI ​​regulation scheme only targets CPU voltage and does not cover multimodal data fusion. In the invention patent with publication number EP3560053B1, the dynamic voltage regulation scheme relies on fixed rules and lacks online learning capabilities. Existing power management integrated circuits lack the capabilities of AI + multimodal collaborative decision-making, 3D packaging microchannel cooling, and protocol adaptive switching.

[0012] Therefore, we have made improvements to this and proposed an intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration and its control method. Summary of the Invention

[0013] (1) The technical problem to be solved by the present invention is that existing power management integrated circuits do not have the functions of AI+ multi-modal collaborative decision-making, 3D packaging microchannel heat dissipation, and protocol adaptive switching.

[0014] (2) Technical solution In order to achieve the above-mentioned purpose of the invention, the present invention provides an intelligent power management integrated circuit with AI dynamic regulation and multimodal collaboration, including: an AI dynamic regulation module, a multi-protocol adaptive charging system and a three-dimensional heterogeneous packaging structure. The AI ​​dynamic regulation module includes a hybrid neural network architecture and a federated learning framework. The multi-protocol adaptive charging system integrates a PD3.1 EPR wired protocol control unit, a Qi v2.0 wireless charging driver module and a dynamic impedance matching circuit. The three-dimensional heterogeneous packaging structure includes TVS and microchannel heat dissipation.

[0015] Among them, the input of the AI ​​dynamic adjustment module includes geographic location data, which is used to predict the impact of network signal strength on radio frequency power consumption.

[0016] The dynamic impedance matching adopts the Smith chart algorithm to optimize the Q value range of the wireless charging coil in real time to 20-80.

[0017] The three-dimensional heterogeneous packaging structure integrates ferroelectric memory, and the radiation resistance of the ferroelectric memory meets the MIL-STD-883 standard.

[0018] It also includes a hierarchical protection mechanism, which includes a self-repairing fuse that automatically restores power supply after the overcurrent event is resolved.

[0019] Among them, the three-dimensional heterogeneous packaging structure adopts a layered structure, which includes a top layer, a middle layer and a bottom layer. The top layer adopts a DC-DC module with a voltage resistance of 30V, the middle layer is a digital control unit, and the bottom layer is a multimodal sensor and ferroelectric memory.

[0020] The DC-DC module uses a GaN HEMT device and has a switching frequency of 5 MHz. The digital control unit is an ARM Cortex-M7 processor or a processor based on the RISC-V architecture, as well as a dedicated neural network accelerator. The multimodal sensor includes a temperature sensor, a current sensor, and a vibration sensor.

[0021] Among them, the microchannel heat dissipation adopts a wavy flow channel, and the wavy flow channel is distributed around the TSV.

[0022] Among them, the ferroelectric memory is located in the upper right quadrant of the bottom layer and is directly connected to the middle-layer NPU through TSV.

[0023] A control method for an intelligent power management integrated circuit with AI dynamic regulation and multi-modal collaboration includes the following steps: S1: Collects device status, environmental data, and user behavior data; S2: Perform noise filtering on the collected data; S3: Input the denoised data into the hybrid neural network model for inference; S4: Generate control strategy based on the reasoning results; S5: Execute the control strategy and adjust and optimize the model through the feedback calibration mechanism.

[0024] (3) Beneficial effects The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration and its control method provided by the present invention have the following beneficial effects: 1. By integrating a highly efficient synchronous Buck-Boost power conversion architecture with an intelligent AI scheduling algorithm, it achieves full-load efficiency exceeding 96% and standby power consumption below 5mW, effectively reducing overall system energy consumption. This addresses the low energy efficiency of traditional PMICs at light loads or idle conditions, and offers precise energy control and overall efficiency advantages over similar products (such as the TI TPS6594).

[0025] 2. Certified to AEC-Q100 Grade 1 (-40°C ~ 125°C) and IEC 62368-1 safety standards, the system can operate stably in extreme temperatures and complex electromagnetic environments, offering high reliability and adaptability to a wide range of industrial applications.

[0026] 3. It reserves an optical communication power supply interface and supports OTA remote upgrade mechanism to achieve continuous adaptation and function expansion under standards such as Wi-Fi 7 and 6G, with the advantages of highly flexible protocol compatibility and system upgrade capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 This is a cross-sectional view of the three-dimensional package of the present invention; Figure 2 This is the AI ​​decision-making flow chart of the present invention; Figure 3 This is a multi-protocol charging efficiency curve diagram of the present invention; Figure 4 This is a schematic diagram of the dynamic impedance matching control logic of the present invention. DETAILED DESCRIPTION

[0029] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0030] Example 1: like Figure 1 、 Figure 2 、 Figure 3 and Figure 4 As shown, this embodiment proposes an intelligent power management integrated circuit with AI dynamic regulation and multi-modal collaboration, including: an AI dynamic regulation module, a multi-protocol adaptive charging system, a three-dimensional heterogeneous packaging structure, and a self-powered safety system; The AI ​​dynamic adjustment module includes a hybrid neural network architecture, a federated learning framework, and scenario examples. The hybrid neural network architecture incorporates LSTM (load time series prediction) and CNN (sensor spatial feature extraction), with a parameter size of less than 50KB and inference latency of less than 2ms. Edge-cloud collaborative training (federated learning framework) supports OTA model updates, reducing cold start time by 80%. The scenario examples predict GPU rendering requirements for gaming scenarios and increase the power supply voltage 10ms in advance to avoid frame rate fluctuations.

[0031] The multi-protocol adaptive charging system includes protocol compatibility and wireless charging optimization, supporting PD 3.1 EPR (28V / 5A), QC 5.0 (100W), and Qi v2.0 (15W), with a dynamic switching time of <5ms. The 4×4 magnetic resonance coil matrix provides 300% spatial freedom and a foreign object detection false trigger rate of <0.01%.

[0032] like Figure 1As shown, the 3D heterogeneous packaging structure utilizes a layered design and microfluidic cooling. The layered structure consists of a top, middle, and bottom layers. The top layer houses a 30V DC-DC module (GaN HEMT device, switching frequency 5MHz), the middle layer houses a digital control unit (ARM Cortex-M7 / RISC-V and dedicated NPU), and the bottom layer houses multimodal sensors (temperature, current, and vibration) and ferroelectric RAM (FeRAM). Copper-based microfluidic channels are integrated into the TSV channels, reducing thermal resistance by 40% and ensuring junction temperature fluctuations of less than ±3°C. Through-silicon vias (TSVs) vertically penetrate the three layers, providing electrical, thermal, and signal interconnects. The top layer connects to the power devices and the bottom layer extends to the FeRAM power supply. Wave-shaped channels surround the TSVs, covering the hotspots (power devices and NPU). Coolant flows from the left inlet, around the TSVs, and out the right outlet. The ferroelectric RAM (FeRAM) is located in the upper right quadrant of the bottom layer and is directly connected to the NPU in the middle layer via TSVs.

[0033] The self-powered safety system includes an energy harvesting module and a graded protection module. The energy harvesting module includes piezoelectric ceramics (vibration energy → 5V / 10mA) and a thermoelectric module (ΔT = 10°C → 3V / 5mA). The graded protection module includes overtemperature (fuse at >125°C), overcurrent (cut off at >5A), short circuit (<10ms response) and self-repairing fuses.

[0034] Example 2: The solution in Example 1 is further introduced below in conjunction with a specific working method, as described below: like Figure 2 As shown, as a preferred implementation, based on the above method, further, the AI ​​dynamic adjustment module input includes geographic location data (GPS / GNSS) for predicting the impact of network signal strength on RF power consumption.

[0035] like Figure 3 As shown in FIG, as a preferred embodiment, based on the above method, the dynamic impedance matching further adopts the Smith chart algorithm to optimize the Q value range of the wireless charging coil in real time to 20-80.

[0036] like Figure 1 As shown in FIG, as a preferred embodiment, based on the above method, a three-dimensional heterogeneous packaging structure is further integrated with ferroelectric memory (FeRAM), and the radiation resistance meets the MIL-STD-883 standard.

[0037] like Figure 4 As shown, as a preferred embodiment, based on the above method, it further includes a hierarchical protection mechanism, which includes a self-repairing fuse that automatically restores power supply after the overcurrent event is resolved.

[0038] Example 3: The following further describes the solutions in Example 1 and Example 2 in conjunction with specific working methods, as described below: Specifically, such as Figure 1 As shown in the figure, the hardware architecture of this intelligent power management integrated circuit with AI dynamic regulation and multimodal collaboration includes a main control unit, a power conversion module and a sensor array. The main control unit includes a processor and memory. The processor has a dual-core architecture (ARM Cortex-M7 and NPU accelerator, supporting INT8 / FP16 mixed precision calculations), and the memory is 128KB SRAM and 1MB FeRAM (radiation-resistant, 10^12 erase and write life).

[0039] like Figure 2 As shown, the power conversion module includes topologies and dynamic impedance matching. The topologies include a synchronous buck-boost (peak efficiency 98%) and a multiphase parallel LDO (quiescent current <1μA). Dynamic impedance matching, based on a Smith chart algorithm, optimizes the Q value of the wireless charging coil in real time (range 20-80). Sensor array inputs include temperature (±0.5°C accuracy), current (0.1mA resolution), and user behavior (GPS / accelerometer).

[0040] The software algorithms of this AI dynamic adjustment and multi-modal collaborative intelligent power management integrated circuit include AI dynamic scheduling and predictive maintenance. AI dynamic scheduling inputs device status, environmental data and user behavior. Device status includes CPU load rate and GPU load rate, environmental data includes temperature and network signal strength, and user behavior includes application usage time.

[0041] Output voltage / frequency combination strategy, power domain switching instructions, and cooling fan speed. AI dynamic scheduling algorithm process: data acquisition → noise filtering (GMM model) → hybrid neural network inference → strategy execution → feedback calibration. Battery health model: Based on coulomb counting and impedance spectroscopy analysis, it predicts capacity decay (error <3%) and provides 30-day advance warning.

[0042] like Figure 3 As shown in the multi-protocol charging efficiency curve, the horizontal and vertical axes are defined as horizontal axis: output power (unit: W, range 5W-140W, logarithmic scale), vertical axis: charging efficiency (unit: %, range 70%-99%).

[0043] Curve data table: like Figure 2 and Figure 4 As shown, the control method of the intelligent power management integrated circuit with AI dynamic regulation and multi-modal collaboration includes the following steps: Collect device status, environmental data, and user behavior data. Device status includes CPU / GPU load rate, environmental data includes temperature and network signal strength, and user behavior data includes application usage time. The collected data was subjected to noise filtering and denoising using Gaussian mixture model (GMM); The denoised data is fed into a hybrid neural network model for inference. The hybrid neural network model includes an LSTM network for load timing prediction and a CNN network for sensor spatial feature extraction. Generate a control strategy based on the inference results, which includes voltage / frequency combination scheme, power domain switching instructions, and cooling fan speed; Execute control strategies and adjust and optimize the model through feedback calibration mechanisms; The control method supports edge-cloud collaboration mechanism, adopts federated learning framework for model training and online update, and supports OTA model update.

[0044] This intelligent power management integrated circuit, featuring AI dynamic regulation and multimodal collaboration, is used in smartphones and new energy vehicles. During a 5G video call on a smartphone, the AI ​​dynamically adjusts the baseband power supply (3.3V to 3.8V) based on signal strength, reducing bit error rates by 45%. It also supports 15W reverse wireless charging with 92% efficiency (compared to 85% with traditional solutions). In new energy vehicle applications, AI optimizes the charge and discharge paths, balancing battery pack SOC variations to less than 1%, extending battery life by 20%. Microchannel heat dissipation minimizes cell temperature differences to less than 2°C, meeting ISO 26262 ASIL-B standards.

[0045] Test data table: The above embodiments are only used to illustrate the present invention, not to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that various combinations, modifications or equivalent substitutions of the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and should be covered by the scope of the claims of the present invention, such as the human-computer interaction applications of wearable products, mobile phone products, industrial control products, PC tablet products, medical products, automotive electronics, smart home and mobile consumer electronics products involved in the present invention.

Claims

1. Intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration, characterized by: include: AI dynamic adjustment module, multi-protocol adaptive charging system and three-dimensional heterogeneous packaging structure. The AI ​​dynamic adjustment module includes a hybrid neural network architecture and a federated learning framework. The multi-protocol adaptive charging system integrates a PD 3.1 EPR wired protocol control unit, a Qi v2.0 wireless charging driver module and a dynamic impedance matching circuit. The three-dimensional heterogeneous packaging structure includes TVS and microchannel heat dissipation.

2. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 1, characterized in that: The AI ​​dynamic adjustment module input includes geographic location data, which is used to predict the impact of network signal strength on radio frequency power consumption.

3. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 1, characterized in that: The dynamic impedance matching uses the Smith chart algorithm to optimize the Q value of the wireless charging coil in real time within the range of 20-80.

4. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 1, characterized in that: The three-dimensional heterogeneous packaging structure integrates ferroelectric memory, and the radiation resistance of the ferroelectric memory meets the MIL-STD-883 standard.

5. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 1, characterized in that: It also includes a hierarchical protection mechanism, which includes a self-repairing fuse that automatically restores power after the overcurrent event is resolved.

6. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 1, characterized in that: The three-dimensional heterogeneous packaging structure adopts a layered structure, which includes a top layer, a middle layer and a bottom layer. The top layer adopts a DC-DC module with a voltage resistance of 30V, the middle layer is a digital control unit, and the bottom layer is a multimodal sensor and a ferroelectric memory.

7. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 6, characterized in that: The DC-DC module uses a GaN HEMT device and has a switching frequency of 5 MHz. The digital control unit is an ARM Cortex-M7 processor or a processor based on the RISC-V architecture, as well as a dedicated neural network accelerator. The multimodal sensor includes a temperature sensor, a current sensor, and a vibration sensor.

8. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 1, characterized in that: Microchannel heat dissipation uses wavy flow channels, which are distributed around the TSV.

9. The intelligent power management integrated circuit with AI dynamic adjustment and multi-modal collaboration according to claim 1, characterized in that: The ferroelectric memory is located in the upper right quadrant of the bottom layer and is directly connected to the middle layer NPU through TSV.

10. A control method for an intelligent power management integrated circuit with AI dynamic adjustment and multi-modal coordination, using the intelligent power management integrated circuit with AI dynamic adjustment and multi-modal coordination as claimed in claim 1, characterized in that: The following steps are involved: S1: Collects device status, environmental data, and user behavior data; S2: Perform noise filtering on the collected data; S3: Input the denoised data into the hybrid neural network model for inference; S4: Generate control strategy based on the reasoning results; S5: Execute the control strategy and adjust and optimize the model through the feedback calibration mechanism.

Images