Predicting optimal parameters for physical design synthesis
View Patent ↗Embodiments of the present disclosure provide enhanced systems and methods for predicting optimal design flow parameters for optimized output targets for physical design synthesis of a given IC design. A Variational Autoencoder (VAE) along with a regression network are trained using a dataset comprising synthesis design construction flows from historical IC designs to provide a training data representation of the dataset constrained to a latent space of the VAE. The system generates feature vectors based on the training data representation of the dataset and updates the feature vectors with initial design characteristics of the given IC design. The system iteratively performs an input gradient search of the updated feature vectors to optimize an objective function of the design targets to identify locally optimal design parameters. The system identifies globally optimal design flow parameters for optimized design targets based on locally optimal design parameters.
1 . A method comprising:
receiving initial design characteristics, design flow parameters, and design targets from a design construction flow for a given integrated circuit (IC) design;
receiving a historical dataset comprising synthesis design construction flows from historical IC designs;
training a Variational Autoencoder (VAE) along with a regression network using the historical dataset to provide a training data representation of the dataset constrained to a latent space of the VAE;
generating feature vectors based on the training data representation of the dataset and updating the feature vectors with the initial design characteristics;
iteratively performing an input gradient search of the updated feature vectors to optimize an objective function of the design targets to identify locally optimal design parameters, wherein the training data samples are constrained to the initial design characteristics; and
obtaining a prediction of global optimal design flow parameters based on the identified locally optimal design parameters for optimized design targets.
2 . The method of claim 1 , further comprises decoding a random set of samples from the training data representation of the dataset to generate feature vectors comprising design characteristics and design flow parameters; and wherein updating the feature vectors with the initial design characteristics comprises replacing design characteristics in the generated feature vectors with the initial design characteristics to provide the updated feature vectors.
3 . The method of claim 1 , wherein iteratively performing the input gradient search further comprises calculating an input gradient for each optimization epoch, applying the input gradient to modify next design flow parameters for a next optimization iteration.
4 . The method of claim 1 , wherein obtaining the prediction of global optimal design flow parameters based on the locally optimal design parameters for the optimized design targets further comprises performing quantization of at least one binary parameter element of the global optimal design flow parameters.
5 . The method of claim 1 , further comprises building the VAE comprising an encoder, the latent space, and a decoder; and wherein the encoder comprises one or more neural network layers with a first plurality nodes on an inner layer and a second plurality of nodes on outer layer; wherein the decoder comprises one or more neural network layers with the second plurality of nodes on a decoder inner layer and the first plurality nodes on a decoder outer layer and wherein the latent space comprises a random sample layer comprising multiple nodes.
6 . The method of claim 1 , wherein obtaining the prediction of global optimal design flow parameters based on the identified locally optimal design parameters for the optimized design targets further comprises sorting the locally optimal design parameters based on the objective function to identify the global optimal design flow parameters.
7 . The method of claim 1 , wherein training the VAE along with the regression network further comprises performing interpolation training of the VAE along with the regression network using combined interpolated vectors and training vectors to minimize a reconstruction error of the training data representation of the dataset.
8 . The method of claim 1 , wherein training the VAE along with the regression network further comprises performing training of the VAE along with the regression network using unsupervised machine learning with a vector of training data features and supervised machine learning with a vector of training data output targets to provide the training data representation of the dataset constrained to the latent space of the VAE.
9 . The method of claim 1 , wherein training the VAE along with the regression network further comprises generating interpolated vectors of the dataset, and combining the interpolated vectors with training vectors to produce an augmented dataset each epoch.
10 . The method of claim 1 , wherein iteratively performing the input gradient search further comprises using hyperparameter values comprising at least one of a regression weight, a reconstruction weight, a number of epochs for training or a training batch size.
11 . The method of claim 1 , wherein the regression network is connected to a neural network layer of the VAE that represents the latent space.
12 . A system, comprising:
a processor; and
a memory, wherein the memory includes a computer program product configured to perform operations for predicting optimal design flow parameters for physical design synthesis of a given integrated circuit (IC) design, the operations comprising:
receiving initial design characteristics, design flow parameters, and design targets from a design construction flow for the given integrated circuit (IC) design;
receiving a historical dataset comprising synthesis design construction flows from historical IC designs;
training a Variational Autoencoder (VAE) along with a regression network using the historical dataset to provide a training data representation of the dataset constrained to a latent space of the VAE;
generating feature vectors based on the training data representation of the dataset and updating the feature vectors with the initial design characteristics;
iteratively performing an input gradient search of the updated feature vectors to optimize an objective function of the design targets to identify locally optimal design parameters, wherein the training data samples are constrained to the initial design characteristics; and
obtaining a prediction of global optimal design flow parameters based on the identified locally optimal design parameters for optimized design targets.
13 . The system of claim 12 , further comprises decoding a random set of samples from the training data representation of the dataset to generate feature vectors comprising design characteristics and design flow parameters; and wherein updating the feature vectors with the initial design characteristics comprises replacing design characteristics in the generated feature vectors with the initial design characteristics to provide the updated feature vectors.
14 . The system of claim 12 , wherein training the VAE along with the regression network further comprises performing training of the VAE along with the regression network using unsupervised machine learning with a vector of training data features and supervised machine learning with a vector of training data output targets to provide the training data representation of the dataset constrained to the latent space of the VAE.
15 . The system of claim 12 , wherein iteratively performing the input gradient search further comprises calculating an input gradient for each optimization epoch, applying the input gradient to modify next design flow parameters for a next optimization epoch.
16 . The system of claim 12 , wherein obtaining the prediction of global optimal design flow parameters further comprises performing quantization of at least one binary parameter element of global optimal design flow parameters.
17 . A computer program product for implementing prediction of optimal design flow parameters for physical design synthesis of a given integrated circuit (IC) design, the computer program product comprising:
a computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to perform an operation comprising:
receiving initial design characteristics, design flow parameters, and design targets from a design construction flow for a given integrated circuit (IC) design;
receiving a historical dataset comprising synthesis design construction flows from historical IC designs;
training a Variational Autoencoder (VAE) along with a regression network using the historical dataset to provide a training data representation of the dataset constrained to a latent space of the VAE;
generating feature vectors based on the training data representation of the dataset and updating the feature vectors with the initial design characteristics;
iteratively performing an input gradient search of the updated feature vectors to optimize an objective function of the design targets to identify locally optimal design parameters, wherein the training data samples are constrained to the initial design characteristics; and
obtaining a prediction of global optimal design flow parameters based on the identified locally optimal design parameters for optimized design targets.
18 . The computer program product of claim 17 , further comprises sorting the locally optimal design parameters based the objective function to obtain the prediction of globally optimal design flow parameters for the given IC design.
19 . The computer program product of claim 17 , wherein training the VAE along with the regression network further comprises generating interpolated vectors of the dataset, and combining the interpolated vectors with training vectors to produce an augmented dataset each epoch.
20 . The computer program product of claim 17 , wherein training the VAE along with the regression network further comprises performing training of the VAE along with the regression network using unsupervised machine learning with a vector of training data features and supervised machine learning with a vector of training data output targets to provide the training data representation of the dataset constrained to the latent space of the VAE.
21 . The computer program product of claim 16 , further comprises decoding a random set of samples from the training data representation of the dataset to generate the feature vectors comprising design characteristics and design flow parameters; and wherein updating the feature vectors with the initial design characteristics comprises replacing design characteristics in the generated feature vectors with the initial design characteristics to provide the updated feature vectors.