Lsi Corporation

Western Digital Technologies, Inc.

Marvell International Limited

The Texas A&m University System

Velodyne Lidar Usa, Inc.

Certicom Corporation

Sandisk Technologies Inc.

Velodyne Lidar, Inc.

Other

Agere Systems Inc.

Kiran Kumar Gunnam

Shaohua Yang

Gwan Seong Choi

Zvonimir Z Bandic

Nedeljko Varnica

Kanke Gao

Nitinkumar Sagarbhai Barot

Gregory Burd

Changyou Xu

Zongwang Li

Yang Han

David S Hall

Anand Kulkarni

Anand Gopalan

Robert Eugeniu Mateescu

Minghai Qin

Arup De

Yuan Xing Lee

Hao Zhong

Nils Graef

Seung-Hwan Song

Yifei Zhang

Jay Scott Fuller

Farshid Rafiee Rad

Rajesh Ramalingam Varadharajan

Roger Pinto

Jay S Fuller

Weijun Tan

Johnson Yen

Viacheslav Anatolyevich Dubeyko

Ishai Ilani

David Declercq

Madhusudan Kalluri

Vikram Varadarajan Kalkunte

Matheus Almeida Ogleari

Erbao Li

Described herein are systems and methods that may efficiently detect multi-return light signals. A light detection and ranging system, such as a LIDAR system, may fire a laser beam that may hit multiple objects with a different distance in one line, causing multi-return light signals to be received by the system. Multi-return detectors may be able to analyze the peak magnitude of a plurality of peaks in the return signals and determine a multitude of peaks, such as the first peak, the last peak and the maximum peak. One embodiment to detect the multi-return light signals may be a multi-return recursive matched filter detector. This detector comprises a matched filter, peak detector, centroid calculation and a zeroing out function. Other embodiments may be based on a maximum finder that algorithmically selects the highest magnitude peaks from samples of the return signal and buffers for regions of interests peaks.

Systems and methods for efficient multi-return light detectors

Certain aspects of the present disclosure provide techniques for partitioning feature maps to improve machine learning model processing. In one aspect, a method, includes partitioning a feature map row-wise into a plurality of feature sub-maps such that: each respective feature sub-map of the plurality of feature sub-maps is defined with respect to a split row determined based on a dense data element count for each row of the feature map; and each feature sub-map of the plurality of feature sub-maps has a same column dimensionality as the feature map; and assigning each of the plurality of feature sub-maps to one of a plurality of tensor compute units and one of a plurality of tensor feature map memory units for processing in parallel.

Just-in-time re-partitioning of feature maps for efficient balancing of compute core workloads

Methods and systems for controlling illumination power of a LIDAR based, three dimensional imaging system based on discrete illumination power tiers are described herein. In one aspect, the illumination intensity of a pulsed beam of illumination light emitted from a LIDAR system is varied in accordance with a set of illumination power tiers based on the difference between a desired and a measured return pulse. In a further aspect, the illumination power tier is selected based on whether an intensity difference exceeds one of a sequence of predetermined, tiered threshold values. In this manner, the intensity of measured return pulses is maintained within a linear range of the analog to digital converter for objects detected over a wide range of distances from the LIDAR system and a wide range of environmental conditions in the optical path between the LIDAR system and the detected object.

LIDAR based distance measurements with tiered power control

A method and system for decoding low density parity check (LDPC) codes. An LDPC code decoder includes LDPC decoding circuitry comprising a Q message generator and a P sum adder array. The Q message generator combines an R message from a previous iteration with a P message to produce a Q message. The P sum adder array adds the P message to a difference of an R message from a current iteration and the R message from the previous iteration to produce an updated P message.

Low density parity check decoder

Described herein are systems and methods that that mitigate avalanche photodiode (APD) blinding and allow for improved accuracy in the detection of a multi-return light signal. A blinding spot may occur due to saturation of a primary APD. The systems and methods include the incorporation of a redundant APD and the utilization of time diversity and space diversity. Detection by the APDs is activated by a bias signal. The redundant APD receives a time delayed bias signal compared to the primary APD. Additionally, the redundant APD is positioned off the main focal plane in order to attenuate an output of the redundant APD. With attenuation, the redundant APD may not saturate and may have a successful detection during the blinding spot of the primary APD. Embodiments may include multiple primary APDs and multiple secondary APDs.

Systems and methods for mitigating avalanche photodiode (APD) blinding

Methods and systems for improved simultaneous localization and mapping based on 3-D LIDAR image data are presented herein. In one aspect, LIDAR image frames are segmented and clustered before feature detection to improve computational efficiency while maintaining both mapping and localization accuracy. Segmentation involves removing redundant data before feature extraction. Clustering involves grouping pixels associated with similar objects together before feature extraction. In another aspect, features are extracted from LIDAR image frames based on a measured optical property associated with each measured point. The pools of feature points comprise a low resolution feature map associated with each image frame. Low resolution feature maps are aggregated over time to generate high resolution feature maps. In another aspect, the location of a LIDAR measurement system in a three dimensional environment is slowly updated based on the high resolution feature maps and quickly updated based on the low resolution feature maps.

Multiple resolution, simultaneous localization and mapping based on 3-D LIDAR measurements

An apparatus includes a tensor compute cluster having a plurality of tensor compute units to process a plurality of sub-feature maps in a machine learning application and a tensor memory cluster having a plurality of tensor feature map memory units to store the plurality of sub-feature maps. The apparatus also includes circuitry to partition an input feature map into the plurality of sub-feature maps such that sparsity in each of the plurality of sub-feature maps satisfies a predetermined threshold, and assign each of the plurality of sub-feature maps to one of the plurality of tensor compute units and one of the plurality of tensor feature map memory units for processing in parallel.

Flexible accelerator for sparse tensors in convolutional neural networks

An apparatus includes a first tensor compute cluster configured to receive first input feature tensors, a second tensor compute cluster configured to receive second input feature tensors more sparse than the first input feature tensors, and a vector accelerator. The apparatus also includes circuitry configured to partition an input feature map into a plurality of input feature tensors based on a compression criteria and assign each of the plurality of input feature tensors to one of the first tensor compute cluster, the second tensor compute cluster, or the vector accelerator based upon at least one of parameters including a sparsity and an optimization parameter.

Flexible accelerator for sparse tensors (FAST) in machine learning

A method and system for decoding low density parity check ('LDPC') codes. An LDPC code decoder includes LDPC decoding circuitry comprising a Q message generator and a P sum adder array. The Q message generator combines an R message from a previous iteration with a P message to produce a Q message. The P sum adder array adds the P message to a difference of an R message from a current iteration and the R message from the previous iteration to produce an updated P message.

A method, circuit, and system for managing wear levelling in non-volatile memory. First, an original physical block address (PBA) for a logical block address (LBA) of a write operation may be received. The original PBA is one of a set of PBAs for data blocks of a non-volatile memory array. Each of these PBAs may be uniquely mapped to a particular LBA using a multistage interconnection network (MIN). A swap PBA may next be determined for the LBA. The swap PBA may be selected from the set of PBAs uniquely mapped using the MIN. Then, the MIN may be configured to map the LBA to the swap PBA. Finally, data of a first data block stored at the original PBA may be swapped with data of a second data block stored at the swap PBA.

Wear leveling in non-volatile memory

A system with a multiplication circuit having a plurality of multipliers is disclosed. Each of the plurality of multipliers is configured to receive a data value and a weight value to generate a product value in a convolution operation of a machine learning application. The system also includes an accumulator configured to receive the product value from each of the plurality of multipliers and a register bank configured to store an output of the convolution operation. The accumulator is further configured to receive a portion of values stored in the register bank and combine the received portion of values with the product values to generate combined values. The register bank is further configured to replace the portion of values with the combined values.

Described herein are systems and methods that may efficiently detect multi-return light signals. A light detection and ranging system, such as a LiDAR system, may fire a laser beam that may hit multiple objects with a different distance in one line, causing multi-return light signals to be received by the system. Multi-return detectors may be able to analyze the peak magnitude of a plurality of peaks in the return signals and determine a multitude of peaks, such as the first peak, the last peak and the maximum peak. One embodiment to detect the multi-return light signals may be a multi-return recursive matched filter detector. This detector comprises a matched filter, peak detector, centroid calculation and a zeroing out function. Other embodiments may be based on a maximum finder that algorithmically selects the highest magnitude peaks from samples of the return signal and buffers for regions of interests peaks.

Growing community of inventors

San Jose, CA, United States of America

Kiran Kumar Gunnam