Neural processing unit for cardiovascular disease prediction artificial neural network model
A neural processing unit includes a controller to receive a compiled machine code of an artificial neural network (ANN) model for predicting cardiovascular disease, the ANN model configured to receive sensing data of ECG, respiration, pulse rate, acceleration, and/or body temperature and to output a probability of disease onset for cerebral infarction, heart failure, and/or ischemic heart disease in a user; an input circuit configured to receive a plurality of input signals corresponding to the ANN model; a processing element (PE) array to perform a calculation of the ANN model; a special function unit (SFU) to perform a special function of calculating the ANN model; and an on-chip memory to store operation data of the ANN model, wherein the controller controls the PE array, the SFU, and the on-chip memory to process the ANN model according to data locality information of the ANN model included in the compiled machine code.
1 . A neural processing unit (NPU) implemented in hardware, the NPU comprising:
a controller configured to receive a compiled machine code of an artificial neural network (ANN) model for predicting cardiovascular disease, the ANN model configured to receive sensing data of at least one of ECG, respiration, pulse rate, acceleration, and body temperature and to output a probability of disease onset for at least one disease among cerebral infarction, heart failure, and ischemic heart disease;
an input circuit configured to receive a plurality of input signals from M heterogeneous sensors corresponding to the ANN model;
a processing element (PE) array formed as an N×M matrix and configured to perform a calculation of the ANN model, the ANN model including computation of at least one fusion layer that combines multiple sensing data to output fused data of the M heterogeneous sensors;
a special function unit (SFU) configured to perform a special function of calculating the ANN model; and
an on-chip memory configured to store operation data of the ANN model,
wherein the controller is further configured to
control the execution of the PE array, the SFU, and the on-chip memory in a sequence optimized based on data locality information of the ANN model included in the compiled machine code, and
selectively delete non-reusable data during the computation of the at least one fusion layer based on the on-chip memory data locality information of the ANN model, and
wherein the controller is further configured such that
a size of data of one of the M heterogeneous sensors is scaled up or scaled down so that data sizes of one axis of the fused data of the M heterogeneous sensors are equal, and
the PE array has a utilization rate that varies according to a size of at least one axis of heterogeneous sensor data.
2 . The NPU of claim 1 , wherein the ANN model is further configured to output the probability of disease onset for the at least one disease by inputting the sensing data of at least two of the ECG, the respiration, the pulse rate, the acceleration, and the body temperature.
3 . The NPU of claim 1 , wherein the ANN model is further configured to input the sensing data of at least two of the ECG, the respiration, the pulse rate, the acceleration, and the body temperature and to output a probability of disease onset for at least two diseases among cerebral infarction, heart failure, and ischemic heart disease.
4 . The NPU of claim 1 , further comprising:
an output unit configured to output a diagnosis result of the at least one disease of the ANN model,
wherein the ANN model is trained to process an inference operation of at least one of classification, semantic segmentation, object detection, and prediction by the PE array.
5 . The NPU of claim 1 , wherein the SFU has at least one function of a skip-connection and a concatenation for fusion of artificial neural networks.
6 . The NPU of claim 1 ,
wherein the controller includes a scheduler, and
wherein the scheduler is configured to control the on-chip memory to preserve specific data stored in the on-chip memory up to a specific operation stage of the ANN model based on the data locality information of the ANN model.
7 . The NPU of claim 1 ,
wherein the PE array includes a plurality of threads, and
wherein the controller is further configured to control the plurality of threads to process a parallel section of the ANN model based on the data locality information of the ANN model.
8 . A neural processing unit (NPU) implemented in hardware, the NPU comprising:
a controller configured to receive a machine code of an artificial neural network (ANN) model for predicting cardiovascular disease, the ANN configured to input sensing data of at least one of ECG, respiration, pulse rate, acceleration, and body temperature and to output a probability of disease onset for at least one disease among cerebral infarction, heart failure, and ischemic heart disease, the sensing data being input from M heterogeneous sensors;
a processing element (PE) array formed as an N×M matrix and configured to perform computation of the ANN model based on the machine code, the ANN model including computation of at least one fusion layer that combines multiple sensing data to output fused data of the M heterogeneous sensors; and
a special function unit (SFU) configured to compute a corresponding special function by receiving a convolution operation value processed by the PE array,
wherein the SFU includes a plurality of function units and is further configured to selectively control at least one of the plurality of function units in a sequence optimized according to data locality information of the ANN model included in the machine code,
wherein the controller is further configured to selectively delete non-reusable data during the computation of the at least one fusion layer based on the memory data locality information of the ANN model, and
wherein the controller is further configured such that
a size of data of one of the M heterogeneous sensors is scaled up or scaled down so that data sizes of one axis of the fused data of the M heterogeneous sensors are equal, and
the PE array has a utilization rate that varies according to a size of at least one axis of heterogeneous sensor data.
9 . The NPU of claim 8 , wherein the plurality of function units are configured in a pipeline structure.
10 . The NPU of claim 8 , wherein the plurality of function units are configured to be selectively activated by the controller.
11 . The NPU of claim 8 , wherein the plurality of function units are configured to be selectively deactivated by the controller.
12 . The NPU of claim 8 , wherein each of the plurality of function units is configured to be selectively clock-gated for each specific operation stage by the controller.
13 . The NPU of claim 8 , wherein each of the plurality of function units is configured to be selectively power-gated for each specific operation stage by the controller.
14 . The NPU of claim 8 , further comprising:
an input unit configured to receive a plurality of input signals corresponding to the ANN model; and
an on-chip memory configured to store computation data of the ANN model.
15 . The NPU of claim 8 , further comprising:
a batch input unit configured to receive a plurality of input signals corresponding to the ANN model in a batch-mode;
an on-chip memory configured to store computation data of the ANN model in the batch-mode; and
an output unit configured to output at least one inference result of the ANN model,
wherein the ANN model is trained to process an inference operation of at least one of classification, semantic segmentation, object detection, and prediction by the PE array in the batch-mode.
16 . A system comprising:
at least one neural processing unit (NPU) implemented in hardware, the NPU including
a controller configured to receive a machine code of an artificial neural network (ANN) model for predicting cardiovascular disease, the ANN model configured to input sensing data of at least one of ECG, respiration, pulse rate, acceleration, and body temperature and to output a probability of disease onset for at least one disease among cerebral infarction, heart failure, and ischemic heart disease,
an input unit configured to receive a plurality of input signals from M heterogeneous sensors corresponding to the ANN model,
a processing element array formed as an N×M matrix and configured to perform a convolution operation including computation of at least one fusion layer that combines multiple sensing data to output fused data of the M heterogeneous sensors, and
an on-chip memory configured to store a result of the convolution operation; and
a memory controller including a memory, the memory controller configured to
receive data locality information of the ANN model, the data locality information of the ANN model predicting successive memory operation requests in an optimized sequence of the at least one neural processing unit, and
cache a next memory operation request to be requested by a corresponding one of the at least one neural processing unit based on the data locality information,
wherein the controller is further configured to selectively delete non-reusable data during the computation of the at least one fusion layer based on the on-chip memory data locality information of the ANN model, and
wherein the controller is further configured such that
a size of data of one of the M heterogeneous sensors is scaled up or scaled down so that data sizes of one axis of the fused data of the M heterogeneous sensors are equal, and
the PE array has a utilization rate that varies according to a size of at least one axis of heterogeneous sensor data.
17 . The system of claim 16 ,
wherein the at least one neural processing unit is plural, and
wherein the machine code of the ANN model input to the controller of each neural processing unit is configured to be processed in parallel in the plurality of neural processing units.
18 . The system of claim 16 ,
wherein the at least one neural processing unit is plural, and
wherein the machine code is compiled for parallel processing in the plurality of neural processing units.
19 . The system of claim 16 ,
wherein the at least one neural processing unit is plural, and
wherein the memory controller is configured to directly control a parallel processing of the plurality of neural processing units.