# US 2021/0174206 A1 · System and method for efficient parallel execution of physical simulations

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0174206 A1
- Document ID: US20210174206A1
- Status: Published application
- Kind: A1
- Application number: 17/111,364
- Filed: 2020-12-03
- Published: 2021-06-10
- Earliest claimed: 2019-12-06
- Assignees: X Development LLC
- Applicants: X Development LLC
- Inventors: Ardavan Oskooi; Christopher Hogan; Alec M. Hammond; Steven G. Johnson
- Web page: https://patentgrep.com/patent/US20210174206A1

## CPC

- G06N 3/08: Learning methods
- G06F 9/5066: {Algorithms for mapping a plurality of inter-dependent sub-tasks onto a plurality of physical CPUs (mappping at compile time, see G06F8/451)}
- G06F 30/20: Design optimisation, verification or simulation (optimisation, verification or simulation of circuit designs G06F30/30)
- G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/09: Supervised learning
- G06F 2209/5019: Workload prediction
- G06N 3/048: Activation functions
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

In some embodiments, a method for optimal parallel execution of a simulation of a design is provided. A computing device extracts one or more features from the design. The computing device provides at least the one or more features as inputs to one or more machine learning models to determine one or more predictions of execution times. The computing device determines an optimum execution architecture based on the one or more predictions of execution times. The computing device distributes portions of the design for simulation based on the optimum execution architecture. In some embodiments, one or more machine learning models are trained to generate outputs for predicting an optimal parallel execution architecture for simulation of a design.

## Claims

### Claim 1

A non-transitory computer-readable medium having logic stored thereon that, in response to execution by one or more processors of a computing device, cause the computing device to perform actions for optimal parallel execution of a simulation of a design, the actions comprising:
extracting, by the computing device, one or more features from the design;
providing, by the computing device, at least the one or more features as inputs to one or more machine learning models to determine one or more predictions of execution times;
determining, by the computing device, an optimum execution architecture based on the one or more predictions of execution times; and
distributing, by the computing device, portions of the design for simulation based on the optimum execution architecture.

### Claim 2 (depends on 1)

The non-transitory computer-readable medium of claim 1, wherein extracting the one or more features from the design includes determining a number of segment types for the design and a total number of segments for the design.

### Claim 3 (depends on 1)

The non-transitory computer-readable medium of claim 1, wherein providing at least the one or more features as inputs to the one or more machine learning models to determine the one or more predictions of execution times includes:
providing at least one proposed number of processing elements as an input to the one or more machine learning models.

### Claim 4 (depends on 3)

The non-transitory computer-readable medium of claim 3, wherein providing at least the one or more features as input to the one or more machine learning models to determine the one or more predictions of execution times includes:
providing the one or more features and the proposed number of processing elements to a first machine learning model model that generates output for predicting computation time; and
providing the one or more features and the proposed number of processing elements to a second machine learning model that generates output for predicting communication time.

### Claim 5 (depends on 4)

The non-transitory computer-readable medium of claim 4, wherein the actions further comprise:
determining a predicted computation time by multiplying the output of the first machine learning model by a total number of segments for the design and dividing by the proposed number of processing elements.

### Claim 6 (depends on 4)

The non-transitory computer-readable medium of claim 4, wherein the actions further comprise:
determining a predicted communication time by multiplying the output of the second machine learning model by a maximum surface area of a portion of the design to be provided to a processing element for simulation.

### Claim 7 (depends on 3)

The non-transitory computer-readable medium of claim 3, wherein determining the optimum execution architecture based on the one or more predictions of execution times includes:
providing different proposed numbers of processing elements as an input to the one or more machine learning models; and
determining a performance plateau based on the outputs of the one or more machine learning models for the different proposed numbers of processing elements.

### Claim 8 (depends on 1)

The non-transitory computer-readable medium of claim 1, wherein distributing portions of the design for simulation based on the optimum execution architecture includes:
instantiating a number of processing elements based on the optimum execution architecture; and
distributing portions of the design to the instantiated processing elements for simulation.

### Claim 9

A method for optimal parallel execution of a simulation of a design, the method comprising:
extracting, by a computing device, one or more features from the design;
providing, by the computing device, at least the one or more features as inputs to one or more machine learning models to determine one or more predictions of execution times;
determining, by the computing device, an optimum execution architecture based on the one or more predictions of execution times; and
distributing, by the computing device, portions of the design for simulation based on the optimum execution architecture.

### Claim 10 (depends on 9)

The method of claim 9, wherein extracting the one or more features from the design includes determining a number of segment types for the design and a total number of segments for the design.

### Claim 11 (depends on 9)

The method of claim 9, wherein providing at least the one or more features as inputs to the one or more machine learning models to determine the one or more predictions of execution times includes:
providing at least one proposed number of processing elements as an input to the one or more machine learning models.

### Claim 12 (depends on 11)

The method of claim 11, wherein providing at least the one or more features as input to the one or more machine learning models to determine the one or more predictions of execution times includes:
providing the one or more features and the proposed number of processing elements to a first machine learning model that generates output for predicting computation time; and
providing the one or more features and the proposed number of processing elements to a second machine learning model that generates output for predicting communication time.

### Claim 13 (depends on 12)

The method of claim 12, wherein the actions further comprise:
determining a predicted computation time by multiplying the output of the first machine learning model by a total number of segments for the design and dividing by the proposed number of processing elements.

### Claim 14 (depends on 12)

The method of claim 12, wherein the actions further comprise:
determining a predicted communication time by multiplying the output of the second machine learning model by a maximum surface area of a portion of the design to be provided to a processing element for simulation.

### Claim 15 (depends on 11)

The method of claim 11, wherein determining the optimum execution architecture based on the one or more predictions of execution times includes:
providing different proposed numbers of processing elements as an input to the one or more machine learning models; and
determining a performance plateau based on the outputs of the one or more machine learning models for the different proposed numbers of processing elements.

### Claim 16 (depends on 9)

The method of claim 9, wherein distributing portions of the design for simulation based on the optimum execution architecture includes:
instantiating a number of processing elements based on the optimum execution architecture; and
distributing portions of the design to the instantiated processing elements for simulation.

### Claim 17

A non-transitory computer-readable medium having logic stored thereon that, in response to execution by one or more processors of a computing device, cause the computing device to perform actions for training a machine learning model to generate outputs for predicting an optimal parallel execution architecture for simulation of a design, the actions comprising:
determining, by the computing device, a set of sample designs;
causing, by the computing device, each sample design to be simulated using at least one hardware configuration;
determining, by the computing device, a set of features for each simulation, wherein the set of features for each simulation includes a number of segment types of the sample design in the simulation, a total number of segments of the sample design in the simulation, and a characteristic of the hardware configuration of the simulation;
training, by the computing device, at least one machine learning model using the features for each simulation and the measured performance of each simulation as training data; and
storing, by the computing device, the at least one machine learning model in a model data store.

### Claim 18 (depends on 17)

The non-transitory computer-readable medium of claim 17, wherein causing each sample design to be simulated using at least one hardware configuration includes:
for a given sample design, causing a first simulation of the given sample design to be executed using a first hardware configuration and causing a second simulation of the given sample design to be executed using a second hardware configuration.

### Claim 19 (depends on 18)

The non-transitory computer-readable medium of claim 18, wherein the characteristic of the hardware configuration is a number of processing elements, wherein the number of processing elements for the first hardware configuration is a first number of processing elements, wherein the number of processing elements for the second hardware configuration is a second number of processing elements, and wherein training the at least one machine learning model using the features for each simulation and the measured performance of each simulation as training data includes:
using the features for the given sample design as a first training data set, and using the measured performance of the first simulation as a label for the first training data set; and
using the features for the given sample design as a second training data set, and using the measured performance of the second simulation as a label for the second training data set.

### Claim 20 (depends on 17)

The non-transitory computer-readable medium of claim 17, wherein training at least one machine learning model using the features for each simulation and the measured performance of each simulation as training data includes:
training a first artificial neural network to generate a value usable to predict computation time; and
training a second artificial neural network to generate a value usable to predict communication time.
