Nvidia Corporation

Matthew Rudolph Fojtik

John Wood Poulton

John M Wilson

Haoxing Ren

Brucek Kurdo Khailany

Carl Thomas Gray

William James Dally

Joel S Emer

Stephen William Keckler

Stephen G Tell

Ilyas Elkin

Nan Jiang

Brian Matthew Zimmer

Rangharajan Venkatesan

Yakun Shao

Benjamin Andrew Keller

Yanqing Zhang

Jason Lavar Clemons

Nathaniel Pinckney

A genetic algorithm is utilized to generate routing candidates to which a reinforcement learning model is applied to correct the design rule constraint violations incrementally. A design rule checker provides feedback on the violations to the reinforcement learning model and the model learns how to fix the violations. A layout device placer based upon a simulated annealing method may also be utilized.

Standard cell layout generation with applied artificial intelligence

The disclosure provides a general solution for determining connections between terminals of various types of circuits using machine learning (ML). A ML method that uses reinforcement learning (RL), such as deep RL, to determine and optimize routing of circuit connections using a game process is provided. In one example a method of determining routing connection includes: (1) receiving a circuit design having known terminal groups, (2) establishing terminal positions for the terminal groups in a routing environment, and (3) determining, by the RL agent, routes of nets between the known terminal groups employing a model that is independent of a number of the nets of the circuit. A method of creating a model for routing nets using RL, a method of employing a game for training a RL agent to determine routing connections, and a RL agent for routing connections of a circuit are also disclosed.

Routing connections in integrated circuits based on reinforcement learning

A distributed deep neural net (DNN) utilizing a distributed, tile-based architecture implemented on a semiconductor package. The package includes multiple chips, each with a central processing element, a global memory buffer, and processing elements. Each processing element includes a weight buffer, an activation buffer, and multiply-accumulate units to combine, in parallel, the weight values and the activation values.

Scalable multi-die deep learning system

This disclosure relates to a circuit comprising a first, second, and third data latch, and an input for a data signal. The first data latch may be configured to sample a delayed version of the data signal in response to a first control signal. The second data latch may be configured to sample the delayed version of the data signal in response to a run clock signal. The run clock signal may be configured to run for a predefined number of clock cycles subsequent to the first control signal. The third data latch may be configured to sample either an output signal of the first data latch or an output signal of the second data latch in response to a second control signal received after the predefined number of clock cycles of the run clock signal.

Signal calibration circuit

A balanced, charge-recycling repeater link is disclosed. The link includes a first set of segments operating in a first voltage domain and a second set of segments operating in a second voltage domain. The link is configured to transmit a first signal over at least one segment in the first set of segments and at least one other segment in the second set of segments. Each segment of the link includes at least one active circuit element configured to charge or discharge one or more corresponding interconnects within the link and a level shifter configured to shift the level of a signal on a last interconnect of the segment from the first voltage domain to the second voltage domain or the second voltage domain to the first voltage domain.

Balanced charge-recycling repeater link

A system, method, and computer program product are provided for a pausible bisynchronous FIFO. Data is written synchronously with a first clock signal of a first clock domain to an entry of a dual-port memory array and an increment signal is generated in the first clock domain. The increment signal is determined to transition near an edge of a second dock signal, where the second clock signal is a pausible clock signal. A next edge of the second clock signal of the second clock domain is delayed and the increment signal to the second clock domain and is transmitted.

Pausible bisynchronous FIFO

A method for operating a latch and a latch circuit are disclosed. The latch circuit comprises a storage sub-circuit, a propagation sub-circuit, and a shared clock-enabled transistor. The storage sub-circuit is configured to capture a level of an input signal when a clock signal transitions from first level to a second level and hold the captured level to generate an output signal while the clock signal is at the second level. The propagation sub-circuit is configured to enable a path through a blocking transistor to the shared clock-enabled supply node to propagate the captured level of the input signal to the storage sub-circuit. The shared clock-enabled transistor is configured to couple the shared clock-enabled supply node to a power supply while the clock signal is at the first level and decouple the shared clock-enabled supply node from the power supply while the clock signal is at the second level.

Growing community of inventors

Raleigh, NC, United States of America

Matthew Rudolph Fojtik