# US 2021/0173983 A1 · Systems and methods for machine learning based fast static thermal solver

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

- Number: US 2021/0173983 A1
- Document ID: US20210173983A1
- Status: Published application
- Kind: A1
- Application number: 16/709,746
- Filed: 2019-12-10
- Published: 2021-06-10
- Earliest claimed: 2019-12-10
- Assignees: Ansys, Inc.
- Applicants: Ansys, Inc.
- Inventors: Norman Chang; Hsiming Pan; Jimin Wen; Deqi Zhu; Wenbo Xia; Akhilesh Kumar; Wen-Tze Chuang; En-Cih Yang; Karthik Srinivasan; Ying-Shiun Li
- Web page: https://patentgrep.com/patent/US20210173983A1

## CPC

- G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G06F 30/398: Design verification or optimisation, e.g. using design rule check [DRC], layout versus schematics [LVS] or finite element methods [FEM] (optical proximity correction [OPC] design processes G03F1/36)
- G06F 2113/18: Chip packaging
- G06F 2119/08: Thermal analysis or thermal optimisation
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Machine assisted systems and methods for enhancing the resolution of an IC thermal profile from a system analysis are described. These systems and methods can use a neural network based predictor, that has been trained to determine a temperature rise across an entire IC. The training of the predictor can include generating a representation of two or more templates identifying different portions of an integrated circuit (IC), each template associated with location parameters to position the template in the IC; performing thermal simulations for each respective template of the IC, each thermal simulation determining an output based on a power pattern of tiles of the respective template, the output indicating a change in temperature of a center tile of the respective template relative to a base temperature of the integrated circuit; and training a neural network. The trained predictor can be used to determine a temperature rise and then can be appended to a system level thermal profile of the IC to generate a detailed thermal profile of the IC.

## Claims

### Claim 1

A non-transitory machine readable medium storing executable program instructions which when executed by a data processing system cause the data processing system to perform a method, the method comprising:
generating a representation of two or more templates identifying different portions of an integrated circuit (IC), each template comprising a plurality of tiles including a center tile, and each template associated with location parameters to position the template in the IC;
performing thermal simulations for each respective template of the IC, each thermal simulation determining an output based on a power pattern of tiles of the respective template, the output indicating a change in temperature of a center tile of the respective template relative to a base temperature of the IC, the power pattern corresponding to a set of power levels powered on the tiles of the respective template for the thermal simulations, each tile of the respective template powered according to one of the set of power levels, each power level selected from a set of predefined power levels; and
training a neural network with a plurality training data generated via the thermal simulations, each training data including the location parameters of one of the templates for inputs to the neural network and including an output of one of the thermal simulations for the one template, the output of the one of the thermal simulations for an output from the neural network.

### Claim 2 (depends on 1)

The medium as in claim 1, wherein the training provides a trained temperature rise predictor.

### Claim 3 (depends on 2)

The medium as in claim 2, wherein the two or more templates comprise a template along an edge of the IC and a template near a center of the IC.

### Claim 4 (depends on 1)

The medium as in claim 1, wherein the two or more templates include three templates on the IC.

### Claim 5 (depends on 1)

The medium as in claim 1, wherein the tiles located outside of each template is powered with an average power level during the thermal simulations.

### Claim 6 (depends on 1)

The medium as in claim 1, wherein the thermal simulations are performed for each template separately and wherein the thermal simulations include computational fluid dynamics simulations or finite element simulations.

### Claim 7 (depends on 1)

The medium as in claim 1, wherein the performing thermal simulations for each respective template of the IC is based on the location parameters and a relationship between a change in temperature relative to a power applied to the IC in the thermal simulations.

### Claim 8 (depends on 7)

The medium as in claim 7, wherein the relationship between a change in temperature relative to power is Theta-JA.

### Claim 9 (depends on 7)

The medium as in claim 7, wherein the relationship between a change in temperature relative to power used in the thermal simulations is varied across the thermal simulations.

### Claim 10 (depends on 1)

The medium as in claim 1, wherein the set of predefined power levels include three or more power levels.

### Claim 11 (depends on 1)

The medium as in claim 1, wherein the tiles of each template located outside a border of the IC is powered with a zero power level during the thermal simulations.

### Claim 12 (depends on 1)

The medium as in claim 1, wherein each template is divided into a center tile group and a ring tile group.

### Claim 13 (depends on 12)

The medium as in claim 12, the tiles in the center tile group are powered with three or more power levels and the tiles in the ring tile group are powered with two power levels to reduce a number of the power pattern during the thermal simulations.

### Claim 14

A non-transitory machine readable medium storing executable program instructions which when executed by a data processing system cause the data processing system to perform a method, the method comprising:
retrieving from memory, a trained neural network model, the neural network model having been trained with a plurality of inputs and an output derived from thermal simulations, the plurality of inputs comprising a relationship between a change in temperature relative to power, a predetermined template location, and a pattern of a plurality of predefined power levels in tiles associated with the template; and
determining, using the retrieved trained neural network model, a change in a temperature of a given tile on an integrated circuit (IC) having a location relative to the predetermined template location, a plurality of inputs to the neural network model including: (1) a selected relationship between a change in temperature relative to a power applied to the IC in the thermal simulations, (2) a selected predetermined location of the template, and (3) a selected pattern of the predefined power level for the template.

### Claim 15 (depends on 14)

The medium as in claim 14, wherein the determining is repeated for each tile on the IC using the retrieved trained neural network model.

### Claim 16 (depends on 14)

The medium as in claim 14, wherein the thermal simulations include computational fluid dynamics (CFD) simulations and finite elements (FE) simulations.

### Claim 17 (depends on 14)

The medium as in claim 14, the method further comprising:
appending a temperature rise map based on the change in the temperature of the given tile to the IC system thermal profile to generate a detailed thermal profile for the IC.

### Claim 18 (depends on 14)

The medium as in claim 14, the method further comprising:
generating a base temperature profile of the IC having a coarse mesh by a system thermal tool based on an initial temperature and a corresponding temperature-dependent power map of the IC;
in response to the generating the base temperature profile of the IC, updating the corresponding temperature-dependent power map of the IC by the system thermal tool;
determining, using the retrieved trained neural network model, a temperature rise of the given tile on the IC based on the updated corresponding temperature-dependent power of the IC;
appending the determined temperature rise of the given tile to the generated base temperature profile of the IC to update the generated base temperature profile of the IC;
updating the generated power map of the IC based on the updated generated base temperature profile of the IC; and
determining, using the retrieved trained neural network model, a subsequent temperature rise of the given tile on the IC based on the updated generated power map of the IC.

### Claim 19 (depends on 14)

The medium as in claim 14, the method further comprising:
iterating a convergence loop based on coarse mesh on the IC to obtain a converged power level for each tile on the IC and a converged thermal profile for each tile on the IC; and
determining, using the retrieved neural network model, a temperature rise of the given tile on an IC having a location relative to the predetermined template location, the plurality of inputs to the neural network model including: (1) the selected relationship between a change in temperature relative to the power applied to the IC in the thermal simulations, (2) the selected predetermined location of the template, and (3) the selected pattern of the converged power level for the template.

### Claim 20 (depends on 14)

The medium as in claim 14, wherein the neural network model is a deep neural network (DNN) model.
