In-situ function parameter search space filtering for machine learning in electronic design automation
A set of parameter values may be generated by a machine learning (ML) model, where the set of parameter values may be used by a black-box function to generate a set of outputs based on a set of inputs. It may be determined whether the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate one or more desired goals. If so, a first response may be provided to the ML model that discourages the ML model from generating sets of parameter values that are similar to the set of parameter values. Otherwise, the set of parameter values may be provided to the black-box function, a second response may be determined based on the set of outputs generated by the black-box function, and the second response may be provided to the ML model.
1 . A method, comprising:
receiving a set of parameter values generated by a machine learning (ML) model, wherein the set of parameter values are used by a black-box function to generate a set of outputs based on a set of inputs;
determining whether the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate one or more desired goals, wherein determining whether the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate the one or more desired goals comprises evaluating one or more surrogate models based on the set of inputs and the one or more desired goals; and
in response to determining, by a processor, that the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate the one or more desired goals, providing a first response to the ML model that discourages the ML model from generating sets of parameter values that are similar to the set of parameter values.
2 . The method of claim 1 , further comprising:
in response to determining that the set of parameter values is not expected to cause the set of outputs generated by the black-box function to violate the one or more desired goals,
providing the set of parameter values to the black-box function,
determining a second response based on the set of outputs generated by the black-box function, and
providing the second response to the ML model.
3 . The method of claim 1 , wherein the first response to the ML model that discourages the ML model from generating the sets of parameter values that are similar to the set of parameter values comprises a fitness value that is greater than a first threshold for desired goals that are minimized or less than a second threshold for desired goals that are maximized, and wherein two sets of parameter values are similar if a distance metric between the two sets of parameter values is less than a third threshold.
4 . The method of claim 1 , wherein the set of inputs includes a circuit design.
5 . The method of claim 4 , wherein a surrogate model in the one or more surrogate models determines if the circuit design violates a foundry design rule.
6 . The method of claim 4 , wherein a surrogate model in the one or more surrogate models determines if the circuit design satisfies a set of assertions.
7 . The method of claim 4 , wherein the one or more desired goals includes ensuring that an area of the circuit design is less than or equal to an area threshold.
8 . The method of claim 4 , wherein the one or more desired goals includes ensuring that a power consumption of the circuit design is less than or equal to a power consumption threshold.
9 . A non-transitory computer-readable medium comprising stored instructions, which when executed by a processor, cause the processor to
receive a set of parameter values generated by a machine learning (ML) model, wherein the set of parameter values are used by a black-box function to generate a set of outputs based on a set of inputs;
determine whether the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate one or more desired goals, wherein determining whether the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate the one or more desired goals comprises evaluating one or more surrogate models based on the set of inputs and the one or more desired goals;
in response to determining that the set of parameter values is expected to cause the set of outputs generated by the black-box function to violate the one or more desired goals, provide a first response to the ML model that discourages the ML model from generating sets of parameter values that are similar to the set of parameter values; and
in response to determining that the set of parameter values is not expected to cause the set of outputs generated by the black-box function to violate the one or more desired goals,
provide the set of parameter values to the black-box function,
determine a second response based on the set of outputs generated by the black-box function, and
provide the second response to the ML model.
10 . The non-transitory computer-readable medium of claim 9 , wherein the first response to the ML model that discourages the ML model from generating the sets of parameter values that are similar to the set of parameter values comprises a fitness value that is greater than a first threshold for desired goals that are minimized or less than a second threshold for desired goals that are maximized, and wherein two sets of parameter values are similar if a distance metric between the two sets of parameter values is less than a third threshold.
11 . The non-transitory computer-readable medium of claim 9 , wherein the set of inputs includes a circuit design.
12 . The non-transitory computer-readable medium of claim 11 , wherein a surrogate model in the one or more surrogate models determines if the circuit design violates a foundry design rule.
13 . The non-transitory computer-readable medium of claim 11 , wherein a surrogate model in the one or more surrogate models determines if the circuit design satisfies a set of assertions.
14 . The non-transitory computer-readable medium of claim 11 , wherein the one or more desired goals includes ensuring that an area of the circuit design is less than or equal to an area threshold.
15 . The non-transitory computer-readable medium of claim 11 , wherein the one or more desired goals includes ensuring that a power consumption of the circuit design is less than or equal to a power consumption threshold.
16 . An apparatus, comprising:
a memory storing instructions; and
a processor, coupled with the memory and to execute the instructions, the instructions when executed causing the processor to:
receive a set of parameter values generated by a machine learning (ML) model, wherein the set of parameter values are used by an electronic design automation (EDA) tool to generate a second circuit design based on a first circuit design;
determine whether the set of parameter values is expected to cause the second circuit design generated by the EDA tool to violate one or more desired goals;
in response to determining that the set of parameter values is expected to cause the second circuit generated by the EDA tool to violate the one or more desired goals, provide a first response to the ML model that discourages the ML model from generating sets of parameter values that are similar to the set of parameter values; and
in response to determining that the set of parameter values is not expected to cause the second circuit design generated by the EDA tool to violate the one or more desired goals,
provide the set of parameter values to the EDA tool,
determine a second response based on the second circuit design generated by the EDA tool, and
provide the second response to the ML model.
17 . The apparatus of claim 16 , wherein the first response to the ML model that discourages the ML model from generating the sets of parameter values that are similar to the set of parameter values comprises a fitness value that is greater than a first threshold for desired goals that are minimized or less than a second threshold for desired goals that are maximized, and wherein two sets of parameter values are similar if a distance metric between the two sets of parameter values is less than a third threshold.
18 . The apparatus of claim 16 , wherein the determining whether the set of parameter values is expected to cause the second circuit design generated by the EDA tool to violate the one or more desired goals comprises evaluating one or more surrogate models based on the first circuit design and the one or more desired goals.
19 . The apparatus of claim 16 , wherein a first goal in the one or more desired goals ensures that an area of the first circuit design is less than or equal to an area threshold, and wherein a second goal in the one or more desired goals ensures that a power consumption of the first circuit design is less than or equal to a power consumption threshold.