A memory-computing integrated chip and a control method

Abstract

This invention discloses an in-memory computing chip and a control method. The method includes: acquiring environmental data surrounding the vehicle; when a preset safety level is met, a first neural network unit outputs the similarity between the input environmental data and multiple environments; based on the similarity of the multiple environments, acquiring the network layers of the hidden layer, the neuron nodes of the hidden layer, multiple neuron nodes of the input layer, multiple neuron nodes of the output layer, and a closed transmission channel; and a second neural network outputting control commands based on the similarity of the multiple environments. This application provides an in-memory computing chip and a control method that solves the technical problem that the control of autonomous vehicles cannot be flexibly adjusted according to the actual changing environment, nor can it meet the control timeliness and comfort requirements of different customers.

Description

Technical Field

[0001] This application relates to the field of semiconductor integrated circuits, and in particular to a memory computing chip and a control method. Background Technology

[0002] In recent years, in order to solve the bottleneck of the traditional von Neumann computing architecture, in-memory computing chip architecture has received widespread attention. Its basic idea is to directly use memory for logical calculations, thereby reducing the amount of data transfer and the transmission distance between memory and processor, reducing power consumption while improving performance.

[0003] Once the existing in-memory computing chip structure for autonomous driving is customized, its circuit structure is fixed and cannot be flexibly adjusted according to the actual changing environment, nor can it meet the control timeliness and comfort requirements of different customers. Summary of the Invention

[0004] This application provides an in-memory computing chip and a control method to solve the technical problems mentioned above, such as the inability of autonomous vehicle control to flexibly adjust according to the actual changing environment and to meet the control timeliness and comfort requirements of different customers.

[0005] In a first aspect, this application provides an in-memory computing chip for use in autonomous vehicles. The in-memory computing chip includes: an acquisition unit for acquiring environmental data surrounding the vehicle; a first neural network unit for outputting similarity scores with multiple environments based on the input environmental data when a preset safety level is met; and a second neural network unit, which includes an input layer, multiple hidden layers, and an output layer. The input layer, output layer, and hidden layer each include multiple neuron nodes. The multiple neuron nodes of the input layer, the multiple neuron nodes of the hidden layer, and the multiple neuron nodes of the output layer are sequentially connected through a transmission channel, which is controlled to open and close by a control switch. The network layer and neuron nodes of the hidden layer, the multiple neuron nodes of the input layer, the multiple neuron nodes of the output layer, and the closed transmission channel are acquired based on the similarity scores of the multiple environments. The second neural network outputs control commands based on the similarity scores of the multiple environments.

[0006] Preferably, training the second neural network includes: using the similarity of multiple environments as the input dataset, the control time as the output dataset, and using control time and comfort as the evaluation function, and using a genetic algorithm to find the network structure of the second neural network. The evaluation function is set as: p = w1A - w2B, where w1 and w2 are weights, A is comfort, and B is control duration.

[0007] Preferably, the first neural network unit, when meeting a preset security level, outputs a similarity score with multiple environments based on the input environmental data; including:

[0008] The first neural network unit has the same network structure as the second neural network;

[0009] The network structure of the first neural network is obtained according to preset conditions, and the first neural network outputs the similarity of the multiple environments based on environmental data.

[0010] Preferably, the first neural network unit, when meeting a preset safety level, outputs the similarity to multiple environments based on the input environmental data; including: a determination unit, which determines the collision risk between the vehicle and surrounding obstacles, and determines the safety level of the vehicle based on different collision risks; and an emergency handling unit, which directly controls the vehicle to perform emergency avoidance when the safety level is less than the preset level.

[0011] Preferably, the in-memory computing chip includes a configuration where the number of transmission channels is greater than the number of channels in the in-memory computing chip.

[0012] The data from the two transmission channels are spliced ​​together to obtain the spliced ​​data to be processed; the processing unit can complete the processing of the spliced ​​data to be processed in one batch.

[0013] This invention provides a control method for autonomous vehicles, which acquires environmental data around the vehicle; when a preset safety level is met, a first neural network unit outputs the similarity between the input environmental data and multiple environments; based on the similarity between the multiple environments, the network layers of the hidden layer and the neuron nodes of the hidden layer, multiple neuron nodes of the input layer, multiple neuron nodes of the output layer, and a closed transmission channel are obtained; and a second neural network outputs control commands based on the similarity between the multiple environments.

[0014] Preferably, when a preset security level is met, the first neural network unit outputs a similarity score with multiple environments based on the input environmental data; including:

[0015] Determine the collision risk between the vehicle and surrounding obstacles, and determine the vehicle's safety level based on different collision risks;

[0016] When the safety level is lower than the preset level, the vehicle will be directly controlled to perform emergency avoidance.

[0017] Preferably, when the number of transmission channels is greater than the number of channels in the in-memory computing chip,

[0018] The data from the two transmission channels are spliced ​​together to obtain the spliced ​​data to be processed; the processing unit can complete the processing of the spliced ​​data to be processed in one batch.

[0019] This invention obtains the network hierarchy and neuron nodes of the hidden layer, as well as the closed transmission channel, by acquiring the similarity of multiple environments. The second neural network outputs control commands based on the input environment similarity. Different structures of the second neural network can be changed according to different environments, thereby achieving the optimal control method, shortening the control time, improving control accuracy, and enhancing environmental adaptability. In addition, the cooperation between the first and second neural networks can further improve the control transmission speed and shorten the control time of autonomous driving. Furthermore, integrating the environmental recognition and decision-making modules in autonomous driving into a single chip, i.e., the first and second neural networks on the same chip, can improve chip reliability and further reduce the risk of damage, thus improving the reliability of autonomous vehicles. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a structural diagram of the in-memory computing chip of the present invention;

[0023] Figure 2 This is another structural diagram of the in-memory computing chip of the present invention;

[0024] Figure 3 This is a flowchart of the control method of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0027] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0028] In a first aspect, this invention discloses an in-memory computing chip for use in autonomous vehicles, such as... Figure 1-2 As shown, the in-memory computing chip includes: an acquisition unit for acquiring environmental data around the vehicle; and a first neural network unit for outputting the similarity to multiple environments based on the input environmental data when a preset safety level is met.

[0029] The second neural network unit includes an input layer, multiple hidden layers, and an output layer. The input layer, output layer, and hidden layer each include multiple neuron nodes. The multiple neuron nodes of the input layer, the multiple neuron nodes of the hidden layer, and the multiple neuron nodes of the output layer are connected sequentially through a transmission channel, which is controlled to be switched on and off by a control switch.

[0030] refer to Figure 1 For example, in this embodiment, the network layers of the hidden layer are set to n layers, the number of neurons in the hidden layer is set to n, the number of neurons in the output layer is set to n, and the above-mentioned neuron nodes are connected through the transmission channel to form an n-layer neural network. The specific structure of the second neural network can be obtained through training, as long as the control accuracy and control time meet the preset conditions.

[0031] The network layers and neuron nodes of the hidden layer, multiple neuron nodes of the input layer, multiple neuron nodes of the output layer, and closed transmission channels are obtained based on the similarity of multiple environments. The second neural network outputs control commands based on the similarity of the multiple environments.

[0032] refer to Figure 2For example, in this embodiment, the network layer of the hidden layer is set to 1 layer, the number of neurons in the hidden layer is set to 5, the number of neurons in the output layer is set to 2, and the number of neurons in the output layer is set to 3. The above-mentioned neurons are connected through the transmission channel to form a 3-layer neural network.

[0033] This invention obtains the network hierarchy and neuron nodes of the hidden layer, as well as the closed transmission channel, by acquiring the similarity of multiple environments. The second neural network outputs control commands based on the input environment similarity. Different structures of the second neural network can be changed according to different environments, thereby achieving the optimal control method, shortening the control time, improving control accuracy, and enhancing environmental adaptability. In addition, the cooperation between the first and second neural networks can further improve the control transmission speed and shorten the control time of autonomous driving. Furthermore, integrating the environmental recognition and decision-making modules in autonomous driving into a single chip, i.e., the first and second neural networks on the same chip, can improve chip reliability and further reduce the risk of damage, thus improving the reliability of autonomous vehicles.

[0034] Preferably, training the second neural network includes: using the similarity of multiple environments as the input dataset, the control time as the output dataset, and using control time and comfort as the evaluation function, and using a genetic algorithm to find the network structure of the second neural network. The evaluation function is set as: p = w1A - w2B, where w1 and w2 are weights, A is comfort, and B is control duration.

[0035] In this field, using genetic algorithms to optimize and find different structures for the second neural network can improve the efficiency of training. During autonomous driving, the driving scenario of the autonomous vehicle can be determined based on the similarity of multiple environments. Different driving scenarios require different control times and levels of comfort. For example, if the autonomous vehicle is in an accident-prone area, the control time needs to be short enough to avoid risks. In environments with fewer surrounding vehicles and relatively smooth roads, control time becomes the most important factor. Therefore, under different safety levels, the goal is to minimize control time and maximize comfort. Different second neural network structures can meet these conditions, while a single neural network structure cannot fully leverage its flexible processing advantages.

[0036] Preferably, the first neural network unit, when meeting a preset security level, outputs a similarity score with multiple environments based on the input environmental data; including:

[0037] The first neural network unit has the same network structure as the second neural network;

[0038] The network structure of the first neural network is obtained according to preset conditions, and the first neural network outputs the similarity of the multiple environments based on environmental data.

[0039] In this field, the input layer, output layer, and hidden layer of the first neural network also employ a variable approach, which will not be elaborated further here. However, in the process of processing environmental data, the processing speed of different neural network structures is obviously inconsistent. Therefore, the preset conditions can be artificially set, requiring a neural network structure with the required accuracy for environmental recognition, or a neural network structure with low power consumption. The specific requirements can be selected according to the actual situation. By setting the first neural network structure as a variable neural network structure, the control time of autonomous vehicles can be further shortened, while also improving the accuracy of environmental recognition.

[0040] Preferably, the first neural network unit, when meeting a preset security level, outputs a similarity score with multiple environments based on the input environmental data; including:

[0041] The judgment unit determines the collision risk between the vehicle and surrounding obstacles, and determines the safety level of the vehicle based on different collision risks.

[0042] The emergency response unit directly controls the vehicle to take emergency evasive action when the safety level is lower than the preset level.

[0043] Preferably, the in-memory computing chip includes a configuration where the number of transmission channels is greater than the number of channels in the in-memory computing chip.

[0044] The data from the two transmission channels are spliced ​​together to obtain the spliced ​​data to be processed; the processing unit can complete the processing of the spliced ​​data to be processed in one batch.

[0045] In theory, the number of channels in a chip can match the number of transmission channels in the second neural network. However, due to manufacturing costs, the number of chips is generally limited to a certain range. The number of transmission channels in the chip is less than that of the input layer and hidden layers. The number of transmission channels in the hidden layers and output layers depends on data processing. For example, during the training of the second neural network, the number of transmission channels in the input layer and hidden layers, and the number of transmission channels in the hidden layers and output layers, might be set to 4. However, in the manufacturing process of an in-memory computing chip, the possible number of transmission channels is 3. This data processing prevents data accumulation and minimizes control time.

[0046] For example, the data to be processed in the transmission channels between the input layer and the hidden layer, and between the hidden layer and the output layer, contains data from four channels: channel 1, channel 2, channel 3, and channel 4. The chip has three input channels. By concatenating the data from channel 1 and channel 2, a fifth channel is obtained. The data from channel 3, channel 4, and channel 5 are then used as the concatenated data to be processed. Thus, the number of channels in the concatenated data to be processed is three. The chip can complete the processing of the first concatenated data in one batch, thus completing the processing of the first data to be processed.

[0047] This invention includes a control method applied to autonomous vehicles, such as... Figure 3 As shown:

[0048] Step S1: Obtain environmental data around the vehicle;

[0049] Step S2: When the preset security level is met, the first neural network unit outputs the similarity with multiple environments based on the input environmental data.

[0050] Step S3: Obtain the network layers of the hidden layer and the neuron nodes of the hidden layer, multiple neuron nodes of the input layer, multiple neuron nodes of the output layer, and the closed transmission channel based on the similarity of multiple environments. The second neural network outputs control commands based on the similarity of the multiple environments.

[0051] This invention obtains the network hierarchy and neuron nodes of the hidden layer, as well as the closed transmission channel, by acquiring the similarity of multiple environments. The second neural network outputs control commands based on the input environment similarity. Different structures of the second neural network can be changed according to different environments, thereby achieving the optimal control method, shortening the control time, improving control accuracy, and enhancing environmental adaptability. In addition, the cooperation between the first and second neural networks can further improve the control transmission speed and shorten the control time of autonomous driving. Furthermore, integrating the environmental recognition and decision-making modules in autonomous driving into a single chip, i.e., the first and second neural networks on the same chip, can improve chip reliability and further reduce the risk of damage, thus improving the reliability of autonomous vehicles.

[0052] Preferably, in step S2, when a preset security level is met, the first neural network unit outputs the similarity to multiple environments based on the input environmental data; including:

[0053] The system assesses the collision risk between the vehicle and surrounding obstacles, determines the vehicle's safety level based on different collision risks, and directly controls the vehicle to perform emergency avoidance when the safety level is lower than the preset level.

[0054] Preferably, when the number of transmission channels is greater than the number of channels in the in-memory computing chip, the data from the two transmission channels are spliced ​​together to obtain the spliced ​​data to be processed; the computing unit can complete the processing of the spliced ​​data to be processed in one processing batch.

[0055] In theory, the number of channels in a chip can match the number of transmission channels in the second neural network. However, due to manufacturing costs, the number of chips is generally limited to a certain range. The number of transmission channels in the chip is less than that of the input layer and hidden layers. The number of transmission channels in the hidden layers and output layers depends on data processing. For example, during the training of the second neural network, the number of transmission channels in the input layer and hidden layers, and the number of transmission channels in the hidden layers and output layers, might be set to 4. However, in the manufacturing process of an in-memory computing chip, the possible number of transmission channels is 3. This data processing prevents data accumulation and minimizes control time.

[0056] For example, the data to be processed in the transmission channels between the input layer and the hidden layer, and between the hidden layer and the output layer, contains data from four channels: channel 1, channel 2, channel 3, and channel 4. The chip has three input channels. By concatenating the data from channel 1 and channel 2, a fifth channel is obtained. The data from channel 3, channel 4, and channel 5 are then used as the concatenated data to be processed. Thus, the number of channels in the concatenated data to be processed is three. The chip can complete the processing of the first concatenated data in one batch, thus completing the processing of the first data to be processed.

[0057] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A memory-compute integrated chip, applied to an autonomous vehicle, characterized in that, The memory-computing integrated chip comprises: An acquisition unit acquires environmental data around a vehicle; A first neural network unit outputs similarity with multiple environments according to input environmental data when a preset safety level is met; A second neural network unit comprises an input layer, multiple hidden layers, and an output layer, the input layer, the hidden layers, and the output layer comprise multiple neuron nodes, the multiple neuron nodes of the input layer, the multiple neuron nodes of the hidden layers, and the multiple neuron nodes of the output layer are sequentially connected through transmission channels, and the transmission channels are controlled by control switches; The network level of the hidden layers and the neuron nodes of the hidden layers are obtained according to the similarity of the multiple environments, the multiple neuron nodes of the input layer, the multiple neuron nodes of the output layer, and the closed transmission channels, and the second neural network outputs a control instruction according to the similarity of the multiple environments; The first neural network and the second neural network are one chip; When training the second neural network, the similarity of the multiple environments is taken as input data set, control time is taken as output data set, control time and comfort are taken as evaluation functions, and a genetic algorithm is used to find the network structure of the second neural network, and the evaluation function is set as: p = w1A-w2B, wherein w1 and w2 are weights, A is comfort, and B is control time length; The first neural network unit outputs similarity with multiple environments according to input environmental data when a preset safety level is met; The first neural network unit has the same network structure as the second neural network; The network structure of the first neural network is obtained according to a preset condition, and the first neural network outputs the similarity of the multiple environments according to environmental data; A determination unit determines the collision risk of a vehicle and surrounding obstacles, and determines the safety level of the vehicle according to different collision risks; An emergency processing unit directly controls the vehicle to perform emergency avoidance when the safety level is less than a preset level; When the number of the transmission channels is greater than the number of channels of the memory-computing integrated chip, the data of two transmission channels are spliced to obtain spliced data to be processed, and an operation unit can complete the processing of the spliced data to be processed through one processing batch.

2. A control method applied to an autonomous vehicle, characterized in that, The method is applied to the memory-computing integrated chip of claim 1, and the method comprises: An acquisition unit acquires environmental data around a vehicle; A first neural network unit outputs similarity with multiple environments according to input environmental data when a preset safety level is met; The network level of the hidden layers and the neuron nodes of the hidden layers are obtained according to the similarity of the multiple environments, the multiple neuron nodes of the input layer, the multiple neuron nodes of the output layer, and the closed transmission channels, and the second neural network outputs a control instruction according to the similarity of the multiple environments.

3. The control method according to claim 2, characterized by The first neural network unit outputs similarity with multiple environments according to input environmental data when a preset safety level is met; ​ The collision risk of the vehicle and the surrounding obstacles is determined, and the safety level of the vehicle is determined according to different collision risks; When the safety level is less than a preset level, the vehicle is directly controlled to perform emergency avoidance.

4. The control method according to claim 2, characterized by The method comprises the steps of: When the number of the transmission channels is greater than the number of the channels of the storage-computing integrated chip, the data of the two transmission channels is spliced to obtain spliced data to be processed; and the operation unit can complete the processing of the spliced data to be processed through one processing batch.

Images