Synopsys, Inc.

Yi-Min Jiang

Sung-Hoon Kwon

Philip Hui-Yuh Tai

Yan Lin

Lin Yuan

Xiang Qiu

Xiang Gao

Balkrishna R Rashingkar

Pedja Raspopovic

Xiang Qui

Philip Hui Yuh Tai

Ping-Hann Peter Wang

Cheng-Hsiang Tsai

Chuan-Yu Wang

Phillip H Tai

Lixin Shao

In one aspect, QoR metrics for different candidate macro placements are estimated using machine learning models. A set of candidate macro placements of hard macros within a circuit design is assessed by estimating a quality metric for each candidate macro placement, as follows. Model-specific estimates of the quality metric are predicted by applying different machine learning models to the candidate macro placements. The different machine learning models are trained using sets of completed macro placements for other circuit designs. The model-specific estimates of the quality metric are combined based on an applicability of (a) the set of macro placements used to train that model to (b) the set of candidate macro placements being evaluated.

Placing hard macros using machine learning predictions trained on different circuit designs

A method to report a phantom object for a structure in a power-and-ground (PG) router is disclosed. The method includes generating the structure of a PG network based on a spec received as input, identifying a violation of a design rule for the structure, and changing the structure to remove the violation of the design rule. The method further includes generating a report of the violation and the changing, generating a phantom object based on the changing, and outputting the report and the phantom object.

Phantom object reporting in a power-and-ground router

A computerized system is disclosed. The computerized system may include one or more processors configured to perform the operations stored in a memory. The operations may include receiving a first circuit design pattern including a DRC violation and generating a first pattern matrix based on the first circuit design pattern, and updating the first circuit design pattern, based on the first pattern matrix, to fix the DRC violation. The operations may also include determining a possibility of a DRC violation-free first circuit design pattern corresponding to the first pattern matrix, and generating a first target label specifying the fixability corresponding to the first pattern matrix based on the determined possibility of the DRC violation-free first circuit design pattern. The first pattern matrix and the first target label may be used as training data to train a machine-learning model to predict fixability of the DRC violation.

Deep learning for fixability prediction of power/ground via DRC violations

Systems and techniques for pattern-based power-and-ground (PG) routing and via rule based via creation are described. A pattern for routing PG wires can be received. Next, an instantiation strategy may be received, wherein the instantiation strategy specifies an area of an integrated circuit (IC) design layout where PG wires based on the pattern are to be instantiated and specifies one or more net identifiers that are to be assigned to the instantiated PG wires. The PG wires can be instantiated in the IC design layout based on the pattern and the instantiation strategy. Additionally, a set of via rules can be received, wherein each via rule specifies a type of via that is to be instantiated at an intersection between two PG wires that are in two different metal layers. Next, one or more vias can be instantiated in the IC design layout based on the set of via rules.

Pattern-based power-and-ground (PG) routing and via creation

A chip layout can include a top-level portion and a set of blocks. The power-and-ground (PG) network for the chip layout can be specified by a set of chip-level PG constraints that is defined using a PG constraint definition language. The set of chip-level PG constraints can be characterized into new sets of PG constraints that correspond to smaller regions of the chip layout, e.g., a set of top-level PG constraints that corresponds to the top-level portion, and a set of block-level PG constraints for each block in the set of blocks. The new sets of PG constraints can then be provided to one or more instances of a PG compiler that executes on one or more processors to create the PG network for the chip layout.

Power-and-ground (PG) network characterization and distributed PG network creation for hierarchical circuit designs

A method of automatically reducing stacked vias while minimizing voltage drop in a power network of an integrated circuit (IC) is provided. In this method, any feasible (i.e. other than connectivity-necessary and uncongested stacked vias) stacked vias of the power network can be virtually removed. If a target voltage drop of the power network is exceeded, then a measurement of the severity of at least a maximum voltage drop on the IC can be updated. After this updating, a set of voltage drop improvement stacked vias can be virtually returned to the power network. The steps of determining whether the target voltage drop is exceeded, updating the severity of the voltage drop at one or more hot spots, and virtually returning the set of additional stacked vias can be repeated until the target voltage drop is not exceeded.

Power network stacked via removal for congestion reduction

A design of an integrated circuit device, in which locations of power wires and memory/logic circuitry are known, is analyzed by at least: identifying intersections of power wires with one another, for power wires that are electrically connected to one another through vias; segmenting power wires, at their intersections; preparing estimates of conductance of vias and wire segments in the form of conductance matrix G; and preparing estimates of current I at each intersection based on power consumed by surrounding circuitry, and current vector 'I' and conductance matrix 'G' are used to solve for voltage drop &#x394;V, in a matrix equation G&#x394;V=I, and the voltage drop is displayed, to allow a human to make changes in the design. Pins of unconnected hard macros are temporarily connected to their closest wires, and current therethrough is included in the estimates.

Power network analyzer for an integrated circuit design

A power pad synthesizer automatically proposes locations of pads that are to carry power in an integrated circuit design. Specifically, a computer is programmed to prepare the plan in at least two stages as follows. In a first stage, a number of pads are proposed around a periphery of the IC design such that an attribute (e.g. maximum voltage drop) satisfies a predetermined condition (e.g. below a user-specified limit that's scaled up by a predetermined amount). In a second stage, the computer automatically identifies a specific location in the design for having an attribute value that satisfies another predetermined condition (e.g. interior location of maximum voltage drop or peripheral location of maximum current), and automatically proposes one or more additional pads at one or more peripheral locations that are identified by a predetermined rule based on the identified specific location (e.g. having the same x-coordinate or the same y-coordinate).

Power pad synthesizer for an integrated circuit design

A plan for a power network for an integrated circuit device is automatically preparing in two stages. In a first stage, a number of simplified plans are prepared on a global scale, without regard to design rule checking constraints and routing blockages. Next, the simplified plans are evaluated to select a plan that conforms to a user-specified limit for an attribute, such as maximum voltage drop. The selected simplified plan, which identifies a total count of power wires and a width of the power wires, is used in a second stage to prepare a more detailed plan that honors the design rule checking constraints and routing blockages. The detailed plan is evaluated to check for conformance with the user-specified limit on the attribute, and if necessary the detailed plan is changed, e.g. by increasing wire width one or more times, to achieve conformance.

Power network synthesizer for an integrated circuit design

A genetic algorithm (GA) based approach to optimize integrated circuit designs for power dissipation. The genetic algorithm optimization process efficiently generates tight lower bounds of the peak power dissipation for a given integrated circuit design. In this approach, the power within a given integrated circuit design circuit is viewed as a function in terms of a set of stimuli to primary inputs of the integrated circuit design. Maximization of the function, and hence, the power dissipation is guided by the genetic algorithm. By repeatedly stimulating the integrated circuit design and measuring the corresponding response, the genetic algorithm process efficiently explores the solution space to obtain a maximization of the function. The genetic algorithm process is implemented within a computer-based EDA (electronic design automation) synthesis system. The EDA synthesis system executes the computer implemented steps of accessing a circuit netlist representing an integrated circuit design to be realized in physical form, defining a function that describes the power with respect to stimulation, maximizing the function by using a genetic algorithm to obtain a set of stimulation inputs which generate a maximum power dissipation, and optimizing power dissipation for the circuit netlist by optimizing those portions of the circuit netlist identified by the set of stimulation inputs.

Growing community of inventors

San Jose, CA, United States of America

Yi-Min Jiang