-
公开(公告)号:US20200050582A1
公开(公告)日:2020-02-13
申请号:US16653578
申请日:2019-10-15
Applicant: Amazon Technologies, Inc.
Inventor: Dana Michelle Vantrease , Ron Diamant
Abstract: A processing element (PE) of a systolic array can perform neural networks computations on two or more data elements of an input data set using the same weight. Thus, two or more output data elements corresponding to an output data set may be generated. Based on the size of the input data set and an input data type, the systolic array can process a single data element or multiple data elements in parallel.
-
公开(公告)号:US10459876B2
公开(公告)日:2019-10-29
申请号:US15885592
申请日:2018-01-31
Applicant: Amazon Technologies, Inc.
Inventor: Dana Michelle Vantrease , Ron Diamant
IPC: G06F15/80 , G06N3/02 , G06F17/16 , G06F15/173
Abstract: A processing element (PE) of a systolic array can perform neural networks computations in parallel on two or more sequential data elements of an input data set using the same weight. Thus, two or more output data elements corresponding to an output data set may be generated in parallel. Based on the size of the input data set and an input data type, the systolic array can process a single data element or multiple data elements in parallel.
-
公开(公告)号:US20190129796A1
公开(公告)日:2019-05-02
申请号:US16160782
申请日:2018-10-15
Applicant: Amazon Technologies, Inc.
Inventor: Ron Diamant , Nafea Bshara , Yaniv Shapira , Guy Nakibly
Abstract: A method for calculating a plurality (M) of redundancy blocks for multiple (N) data blocks of a plurality (D) of words each, the method comprises: receiving the number (M) of redundancy blocks by a calculator that comprises multiple (R) calculation units; configuring the calculator according to M and R; concurrently calculating, if M equals R, by the multiple (R) calculation units of the calculator, R sets of parity vectors, each set includes a plurality (D) of parity vectors; and calculating the plurality (M) of the redundancy blocks based on the R sets of parity vectors.
-
公开(公告)号:US10255151B1
公开(公告)日:2019-04-09
申请号:US15384031
申请日:2016-12-19
Applicant: Amazon Technologies, Inc.
Inventor: Alex Levin , Christopher James BeSerra , Ron Diamant
Abstract: A smart add-in card can be leveraged to perform testing on a host server computer. The add-in card can include an embedded processor and memory. Tests can be downloaded to the add-in card to test a protocol under which the add-in card operates. In a particular example, a PCIe communication bus couples the motherboard to the add-in card and the tests can purposely violate the PCIe specification. The tests can be developed to test conditions that are typically difficult to test without the use of special hardware. However, the smart add-in card can be a simple Network Interface Card (NIC) that resides on the host server computer during normal operation and is used for communication other than security testing. By using the NIC as a testing device, repeatable and reliable testing can be obtained.
-
公开(公告)号:US10212138B1
公开(公告)日:2019-02-19
申请号:US14980664
申请日:2015-12-28
Applicant: AMAZON TECHNOLOGIES, INC.
Inventor: Ron Diamant , Nafea Bshara , Leah Shalev , Erez Izenberg
Abstract: A hardware security accelerator includes a configurable parser that is configured to receive a packet and to extract from the packet headers associated with a set of protocols. The security accelerator also includes a packet type detection unit to determine a type of the packet in response to the set of protocols and to generate a packet type identifier indicative of the type of the packet. A configurable security unit includes a configuration unit and a configurable security engine. The configuration unit configures the configurable security engine according to the type of the packet and to content of at least one of the headers extracted from the packet. The configurable security engine performs security processing of the packet to provide at least one security result.
-
公开(公告)号:US09959227B1
公开(公告)日:2018-05-01
申请号:US14971759
申请日:2015-12-16
Applicant: Amazon Technologies, Inc.
Inventor: Ron Diamant , Georgy Machulsky , Adi Habusha
IPC: G06F13/00 , G06F13/28 , G06F12/0862
CPC classification number: G06F13/28 , G06F12/0862 , G06F2212/6028
Abstract: Apparatus and methods are disclosed herein for reducing I/O latency when accessing data using a direct memory access (DMA) engine with a parser. A DMA descriptor indicating memory buffer location can be stored in cache. A DMA descriptor read command is generated and can include a prefetch command. A descriptor with the indicator can be communicated to the DMA engine in response to the read. A second parser can detect the descriptor communication, parse the descriptor, and can prefetch data from memory to cache while the descriptor is being communicated to the DMA engine and/or parsed by the DMA engine parser. When the DMA engine parses the descriptor, data can be accessed from cache rather than memory, to decrease latency.
-
公开(公告)号:US09858042B1
公开(公告)日:2018-01-02
申请号:US15274585
申请日:2016-09-23
Applicant: AMAZON TECHNOLOGIES, INC.
Inventor: Ron Diamant
CPC classification number: G06F7/588 , H03K3/0315
Abstract: A device includes configurable parallel connected ring oscillators and a finite state machine coupled to the ring oscillators. The finite state machine is configured to cause each of the ring oscillators to operate in an accumulate entropy state for a first period of time and a break phase lock state for a second period of time. When operating in the accumulate entropy state, all of the ring oscillators are in the same configuration. When operating in the break phase lock state, each ring oscillator is in a different configuration than the other ring oscillators.
-
公开(公告)号:US09698819B1
公开(公告)日:2017-07-04
申请号:US15390304
申请日:2016-12-23
Applicant: AMAZON TECHNOLOGIES, INC.
Inventor: Michael Baranchik , Ron Diamant , Muhannad Ghanem , Ori Weber
CPC classification number: H03M7/40 , H03M7/30 , H03M7/405 , H03M7/42 , H03M7/6076
Abstract: A method for generating Huffman codewords to encode a dataset includes selecting a Huffman tree type from a plurality of different Huffman tree types. Each of the Huffman tree types specifies a different range of codeword length in a Huffman tree. A Huffman tree of the selected type is produced by: determining a number of nodes available to be allocated as leaves in each level of the Huffman tree accounting for allocation of leaves in each level of the Huffman tree; allocating nodes to be leaves such that the number of nodes allocated in a given level of the Huffman tree is constrained to be no more than the number of nodes available to be allocated in the given level; and assigning the leaves to symbols of the dataset based an assignment strategy selected from a plurality of assignment strategies to produce symbol codeword information.
-
公开(公告)号:US20170091037A1
公开(公告)日:2017-03-30
申请号:US15282254
申请日:2016-09-30
Applicant: Amazon Technologies, Inc.
Inventor: Ron Diamant , Nafea Bshara , Yaniv Shapira , Guy Nakibly
CPC classification number: G06F11/1092 , G06F11/1076 , G06F11/1096 , G06F11/2094 , G06F2211/1057
Abstract: A method for calculating a plurality (M) of redundancy blocks for multiple (N) data blocks of a plurality (D) of words each, the method comprises: receiving the number (M) of redundancy blocks by a calculator that comprises multiple (R) calculation units; configuring the calculator according to M and R; concurrently calculating, if M equals R, by the multiple (R) calculation units of the calculator, R sets of parity vectors, each set includes a plurality (D) of parity vectors; and calculating the plurality (M) of the redundancy blocks based on the R sets of parity vectors.
-
公开(公告)号:US12242853B1
公开(公告)日:2025-03-04
申请号:US17937335
申请日:2022-09-30
Applicant: Amazon Technologies, Inc.
Inventor: Paul Gilbert Meyer , Ron Diamant , Sundeep Amirineni
Abstract: A compute channel having a compute pipeline of compute stages can be configured using a configuration pipeline with a control table and a datapath table. The control table stores control entries corresponding to respective microoperations, and each control entry includes control information for the compute channel. A datapath table stores datapath configuration entries corresponding to respective microoperations, and each datapath configuration entry has a datapath configuration that includes computational circuit block configurations to configure respective computational circuit blocks in the compute pipeline of the compute channel. Control logic can issue a microoperation to the compute channel by configuring the compute channel according to the control information of the microoperation obtained from the control table, and by inputting the datapath configuration of the microoperation obtained from the datapath table into the configuration pipeline of the compute channel.
-
-
-
-
-
-
-
-
-