Intel Banks on Artificial Intelligence

发布时间:2017-06-26 00:00
来源:EE Times

 Last year, Intel Corp. acquired neural-network hardware maker Nervana and built Nervana’s chip, integrating it with Intel’s own on-processor deep-learning and artificial-intelligence (AI) capabilities. This month, Intel Capital invested in AI startups CognitiveScaleAeye Inc., and Element AI. At the ISC High Performance conferencethis week. Intel fellow Pradeep Dubey outlined the big picture for Intel’s growing AI portfolio.

Intel is investing in AI startups, acquiring others, and blending the mix with its own AI expertise to ensure a leadership position in machine learning, deep learning, and brainlike neural networks based on its AI hardware and software. The company is aiming at all applicable industries, from drug screening with the Xeon Phi to software-defined visualization with its graphics hardware, Dubey said. Intel is also re-architecting its Xeon family for AI by including Altera’s field-programmable gate arrays on chip, he said.

The healthcare industry's first Cognitive Cloud providING person-centric insights to enhance the patient experience, manage the health of populations, and reduce costs. Source: CognitiveScale
The healthcare industry’s first Cognitive Cloud providING person-centric insights to enhance the patient experience, manage the health of populations, and reduce costs.
Source: CognitiveScale

CognitiveScale’s augmented intelligence

The bottom line on Intel’s investment in CognitiveScale is to tap the latter’s expertise, backed by more than 100 patents, on what the startup calls the first industrial-grade “augmented intelligence” software. CognitiveScale defines augmented intelligence as a model of cognition that allows computers to replicate human mental abilities. The goal of its software is to augment human memory, perception, anticipation, problem-solving, and decision-making.

The cognitive software’s deep-learning capabilities are “always on,” allowing it to improve continually by learning from its past experiences of what worked and what did not in six iterative steps: understanding, using a semantic engine that can derive meaning from data and user interactions; interpretation, by representing semantic meaning through both deterministic and probabilistic knowledge graphs; reasoning, using domain-optimized contexts that personalize advice; learning continuously in real-time by comparing historical data with current user interactions; assurance, by maintaining compliance with humanlike principles of responsible risk management; and repeating the process.

AEye develops advanced vision hardware, software and algorithms that act as the eyes and visual cortex of autonomous vehicles. Source: AEye
AEye develops advanced vision hardware, software and algorithms that act as the eyes and visual cortex of autonomous vehicles.
Source: AEye

Element AI’s new world

Perhaps the most enigmatic of the AI startups in which Intel Capital has invested is Element AI, which claims to be conjuring an “AI-First World” that “elevates collective wisdom.” The year-old startup hasn’t detailed its techniques but says it is amalgamating knowledge from entrepreneurs, technology leaders, and academia. The aim, according to Element AI, is to translate the world’s most important AI research into “transformative business applications” that will be personalized for the needs of particular companies. Element AI has coined the term called AI-as-a-Service (AIaaS) to describe its technology, which it plans to keep in-house.

Jean-Francois Gagne's map of the AI ecosystem in Canada is a constant work in progress. In fact, each time he presents his graphic to an audience, he finds more names to add to the collection of dozens of logos, from research labs to incubators to startups. Click here for larger image
Jean-Francois Gagne’s map of the AI ecosystem in Canada is a constant work in progress. In fact, each time he presents his graphic to an audience, he finds more names to add to the collection of dozens of logos, from research labs to incubators to startups. 
Click here for larger image

AEye’s vision

AEye, on the other hand, develops vision algorithms for vehicles, plus the hardware and software to execute them optimally. Its patented approach emulates human eyes and the brain’s visual cortex to make autonomous vehicles visually “smart.” Using solid-state LiDAR scanners, AEye claims to have achieved a variety of breakthroughs in intelligent sensing and perception technology.


 Intel and Facebookare working together on a new cheaper Artificial Intelligence (AI) chip that will help companies with high workload demands.At the CES 2019 here on Monday, Intel announced "Nervana Neural Network Processor for Inference" (NNP-I)."This new class of chip is dedicated to accelerating inference for companies with high workload demands and is expected to go into production this year," Intel said in a statement.Facebook is also one of Intel's development partners on the NNP-I.Navin Shenoy, Intel Executive Vice President in the Data Centre Group, announced that the NNP-I will go into production this year.The new "inference" AI chip would help Facebook and others deploy machine learning more efficiently and cheaply.Intel began its AI chip development after acquiring Nervana Systems in 2016.Intel also announced that with Alibaba, it is developing athlete tracking technology powered by AI that is aimed to be deployed at the Olympic Games 2020 and beyond.The technology uses existing and upcoming Intel hardware and Alibaba cloud computing technology to power a cutting-edge deep learning application that extracts 3D forms of athletes in training or competition."This technology has incredible potential as an athlete training tool and is expected to be a game-changer for the way fans experience the Games, creating an entirely new way for broadcasters to analyse, dissect and re-examine highlights during instant replays," explained Shenoy.Intel and Alibaba, together with partners, aim to deliver the first AI-powered 3D athlete tracking during the Olympic Games Tokyo 2020."We are proud to partner with Intel on the first-ever AI-powered 3D athlete tracking technology where Alibaba contributes its best-in-class cloud computing capability and algorithmic design," said Chris Tung, CMO, Alibaba Group. 
2019-01-09 00:00 阅读量:891
The Israeli parliamentary finance committee approved a $185 million grant to Intel in return for meeting job creation targets and local contract guarantees.Last May, Intel announced it would spend $5 billion over two years to upgrade its Fab 28 in Kiryat Gat, Israel, from 22nm to 10nm production technology.Israel's grant is conditional on Intel meeting its already announced commitment to hire 250 new staff at the fab, and on awarding contracts worth around $560 million to local suppliers.Earlier this month, Ann Kelleher, Intel’s senior vice president and general manager of manufacturing and operations, said the company was planning for manufacturing site expansions in Oregon, Ireland and Israel, with multi-year construction activities expected to start in 2019.In a blog post, Kelleher said, “Having additional fab space at-the-ready will help us respond more quickly to upticks in the market and enables us to reduce our time to increased supply by up to roughly 60%. In the weeks and months ahead, we will be working through discussions and permitting with local governments and communities.”Intel's Fab 28 in Kiryat Gat, Israel.Kelleher said it was part of the company’s strategy to prepare the company’s global manufacturing network for flexibility and responsiveness to demand. As part of this, the company is spending to expand its 14nm manufacturing capacity, made progress on the previously announced schedule for the Fab 42 fit-out in Arizona, and located development of a new generation of storage and memory technology at its manufacturing plant in New Mexico.Kelleher also said that Intel would also supplement its own manufacturing capability with selective use of foundries for certain technologies "where it makes sense for the business." The company had already been doing this but will do so more as it aims to address a broader set of customers beyond the PC and into a $300 billion market for silicon in cars, phones, and artificial intelligence (AI) based products.
2019-01-04 00:00 阅读量:870
The world’s two largest semiconductor companies both presented new technologies for embedded MRAM in logic chip manufacturing processes last week at the 64th International Electron Devices Meeting (IEDM) here.Intel (Santa Clara, California) described the key features of spin-transfer torque (STT)-MRAM–based non-volatile memory into its 22FFL process, calling it “the first FinFET-based MRAM technology.” Describing the technology as “production-ready,” Intel did not name any foundry customers for the process, but multiple sources said that it is already being used in products now shipping.Samsung (Seoul), meanwhile, described STT — MRAM in a 28-nm FDSOI platform. STT-MRAM is regarded as the best MRAM technology in terms of scalability, shape dependence, and magnetic scalability.MRAM technology has been in development since the 1990s but has yet to achieve widespread commercial success. “I think it’s time we show something manufacturable and something commercial,” said Yoon Jong Song, a principal engineer in Samsung’s R&D center and the lead author of the company’s IEDM paper.In addition to being seen as a promising candidate for standalone devices to replace memory chip stalwarts DRAM and NAND flash — which are facing serious scaling challenges as the industry moves to smaller nodes — MRAM, which is a non-volatile memory, is appealing as an embedded technology replacement for flash and embedded SRAM because of its fast read/write times, high endurance, and strong retention. Embedded MRAM is considered especially promising for applications such as internet-of-things (IoT) devices.An electron microscope image showing the cross-section of the MTJ array, which is embedded between Metal 2 and Metal 4 in Intel’s 22-nm FinFET logic process. (Source: IEDM/Intel)Globalfoundries has, since last year, been offering embedded MRAM on its 22FDX 22-nm FD-SOI process. But Jim Handy, principal analyst with Objective Analysis, said that he is not aware of any commercial products shipping the Globalfoundries’ embedded MRAM technology.“The reason why nobody has picked it up is because of the fact that they would have to add new materials,” he said.But embedded MRAM is gaining greater consideration as manufacturing costs come down and other memory technologies face scalability challenges. “The big deal is that, with new process technologies, the size of the SRAM cell isn’t shrinking with the rest of the process, so MRAM becomes increasingly appealing,” said Handy.In its paper, Intel said that its embedded MRAM technology achieves 10-year retention at 200°Celsius and endurance of more than 106 switching cycles. The technology uses a 216 × 225 mm 1T-1R memory cell.Samsung, meanwhile, described its 8-Mb MRAM with endurance of 106 cycles and retention of 10 years.Song said that the Samsung technology will be initially available for IoT applications, adding that the reliability has to improve before it can be used in automotive and industrial applications. “We have successfully transferred the technology from lab to fab and will be moving it to market in the near future.”
2018-12-13 00:00 阅读量:1740
STMicroelectronics says it is the first chip maker to develop and market a certified solution for seamless global, ultra-low-power, and long-range wireless IoT connectivity enabling the Monarch worldwide tracking and positioning service from Sigfox, a lead IoT service provider.ST’s solution lets users create region-independent smart objects that connect automatically to the local Sigfox network anywhere in the world, empowering inter-regional mobility, geolocation, and asset tracking without relying on more expensive GPS or GNSS positioning devices. These could include smart-baggage products that aid tracking in airports or transport hubs, or innovations for supply-chain management and air or rail transportation in the industrial asset-management market, such as smart pallets. Regional independence also allows makers of connected smart objects such as consumer or commercial IoT devices to standardise products for multiple export markets, simplifying manufacturing and logistics.ST is providing a complete Software Development Kit (SDK) that runs on STM32, for Sigfox Monarch networking, supported by development kits, reference designs, and tools that accelerate project completion.The fully certified Sigfox Monarch solution is based on ST’s S2-LP ultra-low-power, long-range, sub-1GHz radio, which is automatically tuned on the local regional Sigfox frequency band, across all relevant worldwide zones (RC1 to RC6), enabling seamless connectivity to the Global Sigfox network and geolocation services.For processing demanding applications, the ARM Cortex-M4 based devices are said to enable efficient data pre-processing and localised AI, reducing network traffic requirements.To jump-start new product development, ST’s S2-LP radio is also available for the STM32 Open Development Environment (ODE) with X-NUCLEO-S2868A1/-S2915A1 (upcoming in late Q4’18) expansion boards and X-CUBE-SFXS2LP1 Sigfox Ready software expansion pack.The SDK also supports a dual-radio reference solution powered by the S2-LP and BlueNRG-2 Bluetooth low energy SoC, design to offer easy in-field provisioning, maintenance and configuration of the device through a convenient smartphone app.The design can be enhanced with the STSAFE secure element for robust cyber-protection and with ST’s comprehensive industrial portfolio of motion and environmental MEMS sensors.
2018-10-31 00:00 阅读量:1609
  • 一周热料
  • 紧缺物料秒杀
型号 品牌 询价
PCA9306DCUR Texas Instruments
TL431ACLPR Texas Instruments
TXB0108PWR Texas Instruments
TPS61021ADSGR Texas Instruments
CD74HC4051QPWRQ1 Texas Instruments
TPIC6C595DR Texas Instruments
型号 品牌 抢购
TPS63050YFFR Texas Instruments
TPS61021ADSGR Texas Instruments
TPS61256YFFR Texas Instruments
ULQ2003AQDRQ1 Texas Instruments
TPS5430DDAR Texas Instruments
TXS0104EPWR Texas Instruments
AMEYA360商城(上线于2011年,现 有超过3500家优质供应商,收录600万种产品型号数据,100 多万种元器件库存可供选购,产品覆盖MCU+存储器+电源芯 片+IGBT+MOS管+运放+射频蓝牙+传感器+电阻电容电感+ 连接器等多个领域,平台主营业务涵盖电子元器件现货销售、 BOM配单及提供产品配套资料等,为广大客户提供一站式购 销服务。