Strategic implications of Chinese/US AI development

Discussion in 'Strategic Defense' started by shifty_ginosaji, Nov 13, 2018.

  1. tidalwave
    Offline

    tidalwave Senior Member
    Registered Member

    Joined:
    Feb 10, 2015
    Messages:
    1,349
    Likes Received:
    1,269
    Huawei is the most influenctial technology company in China

    It would help greatly if it phase out all US components in a few years and do everything in house or get some domestic supplier.

    If it can be a component supplier for wide range of products like Broadcom or Samsung, then it would change the landscape greatly.

    Alot of component supplier in China have problems of slow adoption of their products.

    Since Huawei is end product or system maker it will use the components it developed and that would help greatly.

    It's single greatest darkhorse in making China self independent.
     
    bluewater2012 likes this.
  2. tidalwave
    Offline

    tidalwave Senior Member
    Registered Member

    Joined:
    Feb 10, 2015
    Messages:
    1,349
    Likes Received:
    1,269
    [​IMG]
    本次光博会,集团展出了5G、智慧应用和光电子的最新产品和解决方案。其中,最为瞩目的光电子展出内容,由光迅科技和国家光电子创新中心联合参展。本届光博会光迅主推5G前传、中传光模块解决方案、数据中心等系列产品及方案,重点展示25G SFP28 LR光模块、100G QSFP28 4WDM光模块、50G PAM4 DWDM QSFP28光模块、25G SFP28 Bidi、100G SFP-DD AOC、 400G OSFP SR8等性能业界领先的拳头产品。国家光电子创新中心通过构建基于网络协同的平台共建新模式,与法国III-V实验室以及美国、德国等国外数家知名研究机构开展协同联合攻关,将在展会现场隆重推出国内首款商用100G/200G硅光相干收发芯片和器件

    Yes, 100G/200G Silicon Photonic Transceiver. Soon Huawei and ZTE won't need those from US supplier
     
    Equation and N00813 like this.
  3. Jura
    Offline

    Jura General

    Top Poster Of Month

    Joined:
    Jun 23, 2013
    Messages:
    22,097
    Likes Received:
    27,066
    google tranlation of:
    is
    "At this fair, the group exhibited the latest products and solutions for 5G, smart applications and optoelectronics. Among them, the most eye-catching photonics exhibition content was jointly exhibited by Guangxun Technology and National Optoelectronics Innovation Center. This year's Optical Expo Guangxun mainly promotes 5G pre-transmission, medium-transmitting optical module solutions, data center and other products and solutions, focusing on 25G SFP28 LR optical module, 100G QSFP28 4WDM optical module, 50G PAM4 DWDM QSFP28 optical module, 25G SFP28 Bidi, 100G SFP-DD AOC, 400G OSFP SR8 and other industry-leading products. The National Optoelectronics Innovation Center will build a new model based on a network-based collaboration platform, and cooperate with several French-based research institutes such as the French III-V laboratory and the United States, Germany, etc., and will launch the first domestic commercial 100G at the exhibition site. /200G silicon optical coherent transceiver chip and device"
     
    Equation, zealotaiur485 and N00813 like this.
  4. Anlsvrthng
    Offline

    Anlsvrthng Senior Member
    Registered Member

    Joined:
    Sep 27, 2017
    Messages:
    1,168
    Likes Received:
    270
    This is fact :
    This is prediction :
    Question, do you know what is the cost benefit of the 5G, and what 5G the Huawei supply ?
    And what will be the cost of the true 5G handset?

    Do you remember the discussion about AESA radars, stealth etc ?

    So, the 5G handsets has to have a mm wavelength AESA beam-forming transmitter / receiver.

    So, clever guys will be able to use the "real" 5G handsets as AESA radars : )

    But, the price will reflect these capabilities.
     
  5. Hendrik_2000
    Online

    Hendrik_2000 Brigadier

    Joined:
    Dec 20, 2006
    Messages:
    7,634
    Likes Received:
    26,359
    They will be someday when China's FAB can produced more chips at higher quality, lower cost. There will be incentive for Huawei and I bet even prodding from the central government But chip design were never bee a problem Now Huawei goes after server Chip I posted the article in trade war Here is another version
    via Emperor

    JANUARY 7, 2019 / 10:19 AM / UPDATED 2 HOURS AGO
    Huawei launches server chipset as China pushes to cut reliance on imports
    Sijia Jiang
    4 MIN READ

    HONG KONG (Reuters) - Huawei Technologies Co Ltd [HWT.UL] on Monday launched a new chipset for use in servers, at a time when China is pushing to enhance its chip-making capabilities and reduce its heavy reliance on imports, especially from the United States.

    Huawei, which gets the bulk of its revenue from the sale of telecommunications equipment and smartphones, is seeking growth avenues in cloud computing and enterprise services as its equipment business comes under increased scrutiny in the West amid worries about Chinese government influence over the firm.

    Huawei has repeatedly denied any such influence.

    Chinese firms are also seeking to minimize the impact of a trade dispute that has seen China and the United States slap tariffs on each other’s technology imports.

    For Huawei, the launch of the chipset - called the Kunpeng 920 and designed by subsidiary HiSilicon - boosts its credentials as a semiconductor designer, although the company said it had no intention of becoming solely a chip firm.

    “It is part of our system solution and cloud servicing for clients ... We will never make our chipset business a standalone business,” said Ai Wei, who is in charge of strategic planning for Huawei’s chipsets and hardware technology.


    The Shenzhen-based company already makes the Kirin series of smartphone chips used in its high-end phones, and the Ascend series of chipsets for artificial intelligence computing launched in October.

    It said its latest 7 nanometre, 64-core central processing unit (CPU) would provide much higher computing performance for data centers and slash power consumption. It is based on the architecture of British chip design firm ARM - owned by Japan’s SoftBank Group Corp (9984.T) - which is seeking to challenge the dominance in server CPUs of U.S. maker Intel Corp (INTC.O).

    Huawei aims to “drive the development of the ARM ecosystem”, said Chief Marketing Officer William Xu. He said the chip has “unique advantages in performance and power consumption”.

    Xu also said Huawei will continue its “long-term strategic partnership” with Intel.

    Huawei’s new ARM-based CPU is not a competitor to the U.S. company’s x86 CPUs and servers, but complementary, Xu added.

    Redfox Qiu, president of the intelligent computing business department at Huawei, said the company shipped 900,000 units of servers in 2018, versus 77,000 in 2012 when it started.

    Huawei was seeing “good momentum for the server business in Europe and Asia Pacific” and expects the contribution from its international business to continue to rise, Qiu added.

    Huawei on Monday also released its TaiShan series of servers powered by the new chipset, built for big data, distributed storage and ARM native applications.

    The firm founded chip designer HiSilicon in 2004 to help reduce its reliance on imports.

    In modem chips, Huawei internally sources 54 percent of those in its own devices, with 22 percent coming from Qualcomm Inc (QCOM.O) and the remainder from elsewhere, evidence presented at an antitrust trial for Qualcomm showed.
     
    Equation, zealotaiur485 and N00813 like this.
  6. Hendrik_2000
    Online

    Hendrik_2000 Brigadier

    Joined:
    Dec 20, 2006
    Messages:
    7,634
    Likes Received:
    26,359
    Now for radar and weapon enthusiast here is more exciting new China has just produced anti radiation DSP via xyz. they become immune to EMP very important for jet fighter, ship radar and electronic

    High-performance anti-radiation DSP chip "即墨芯" released
    数字信号处理芯片 “即墨芯”正式发布
    At 11:03 on January 6, 2019 , Qingdao government network


      青岛日报讯 昨日,由国家专用集成电路设计工程技术研究中心(青岛)研发的数字信号处理芯片“即墨芯”在即墨经济开发区海景花苑大酒店正式发布。

      作为现代信息产业的基础和核心产业之一,集成电路产业是关系国民经济和社会发展全局的基础性、先导性和战略性产业。“即墨芯”为抗辐射加固数字信号处理芯片,相比国际同类产品性能提高了20%,具备了国际先进水平,填补了青岛地区在高性能数字信号处理芯片领域的空白。

      据了解,中国科学院自动化研究所国家专用集成电路设计工程技术研究中心在数字信号处理芯片设计领域处于国内领先地位。2016年,该中心与即墨区人民政府合作,成立国家专用集成电路设计工程技术研究中心(青岛),致力于高性能数字信号处理芯片的研发和产业化推广。“即墨芯”的成功研发,将为空间设备的智能化提供强劲动力,也将为保障核心元器件自主可控贡献力量。
    关键字 : 即墨芯片数字信号处理
    我要反馈


      Qingdao Daily News yesterday, the National IC Design Engineering Technology Research Center dedicated (Qingdao) research and development of digital signal processing chip "Jimo core" released in Jimo Economic Development Zone, Sea View Garden Hotel.

      As one of the modern information industry base and core industry, IC industry is overall national economic and social development of basic, guiding and strategic industries. "Jimo core" anti-radiation hardened digital signal processing chip, the performance compared to similar products increased by 20%, with the international advanced level, to fill the gaps in Qingdao in high-performance digital signal processing chip area.

      It is understood that the National ASIC Design Engineering Research Center Institute of Automation, Chinese Academy of Sciences in the domestic leader in digital signal processing chip design. 2016, in cooperation with the Center Jimo District People's Government, the establishment of the National Engineering Research Center for IC Design special (Qingdao), is committed to high-performance digital signal processing chip development and industrialization promotion. "Jimo core" successfully developed, will provide a strong impetus intelligent space equipment, will also protect the core components contribute to self-control.

    https://news.sina.com.cn/c/2019-01-06/doc-ihqfskcn4484772.shtml
     
    zealotaiur485 and N00813 like this.
  7. tidalwave
    Offline

    tidalwave Senior Member
    Registered Member

    Joined:
    Feb 10, 2015
    Messages:
    1,349
    Likes Received:
    1,269
    Photonic Chip is 1000 times faster than traditional chip but the manufacturing process of photonic chip leveraged from traditional silicon process. Silicon process is used to constructed wave guide, wave ring structure for light to travel.

    130nm process is used for the photonic chip production. Old process is good enough, whereas the traditional chip is using 7nm currently, very costly.

    Photonic chip is better fitted for AI. Why?? Because traditional chip is very hard to mimic neuron whereas Photonic chip is better at mimic neuron structure!!

    Tsinghua, Peking university research group came out with photonic chip for AI.

    我国首个光子人工智能芯片项目正式启动!光子芯片原理是什么?为什么更适合于AI?


    光子人工智能芯片。光子芯片的计算速度为电子芯片的1000倍,但功耗仅为其百分之一。

    算力是传统电子人工智能芯片的1000倍,但功耗只有其百分之一,低延迟还抗电磁干扰,由清华、北大、北交大等高校博士生创业研发的光子人工智能芯片,在技术上实现不少突破,未来可广泛应用于手机、自动驾驶、智能机器人、无人机等领域。近日,该光子人工智能芯片项目落户顺义,将这项新技术推向了台前。

    “芯片的设计、加工、封装、测试全部在国内完成,摆脱了对国外高制程光刻机的依赖,是我国在芯片领域换道超车的核心技术。”研究团队负责人白冰说。

    具高速率、低功耗优势

    落地顺义的光子人工智能芯片出自一个由清华、北大、北交大等多所高校的在校博士生组成的创业团队,该团队是全国第一个,也是全球第二个光子人工智能芯片研究团队。

    团队负责人、光子芯片的研发者之一——白冰,目前正在北京交通大学通信与信息系统专业攻读博士学位。作为国内第一个研究光子计算的团队,白冰介绍,光子芯片具有低延迟、抗电磁干扰等优势,计算能力是传统芯片的三个数量级,功耗却只有传统芯片的百分之一。
    [​IMG]
    全流程可在国内完成

    对于光子人工智能芯片发展的意义,白冰说,国内电子芯片设计领域能力很强,但在核心加工环节有一个很强的技术壁垒,需依赖国外的高制程光刻机,在成本等多个方面都会受限。光子人工智能芯片的生产过程自主可控,全流程可在国内完成,采用国内130nm微电子工艺加工完成,摆脱了对于国外高制程光刻机的依赖,无需在工艺制程上进行追赶。

    “芯片的设计、加工、封装、测试全部在国内完成,摆脱了对国外高制程光刻机的依赖,所以说是我国在芯片领域换道超车的核心技术。”白冰说。

    白冰表示,未来芯片主要还是针对人工智能领域的应用与发展。目前光子人工智能芯片的产品部署主要集中于设备端,预计于2022年将光子芯片运用到云端。

    记者了解到,第三代半导体是北京市高精尖产业的重要内容,也是顺义确定发展的三大创新型产业集群之一。当前,顺义正布局全产业链,7.1万平方米的第三代半导体材料及应用联合创新基地已于去年12月竣工。目前,中关村科技园区顺义园管理委员会已与其团队签约,项目正式落户顺义。

    据介绍,上海也已开始光子芯片的布局,计划到2021年建成全国硅光子芯片研发和中试基地,到2025年量产平台实现芯片批量供货,成为国际知名硅光子的研发、制造基地。

    释疑1 什么是光子人工智能芯片?

    白冰介绍,光子人工智能芯片是指采用硅基光子集成技术,让光提供算力,为人工智能应用提供高性能的硬件支持。第一个层面是“人工智能芯片”。如果一个芯片要跑得非常快、非常省电,一定是芯片的物理结构跟软件高度匹配,这样才能达到一个比较高的效率。包括现在的人脸识别、自动驾驶、安防监控、AI金融、AI医疗等,实际上都是一种人工智能算法,要设计一款芯片结构跟其特征匹配,这就是人工智能芯片。

    第二个层面是“光子”。已有的人工智能芯片都是电子芯片,电子芯片在计算速度和功耗方面会有瓶颈。光子人工智能芯片是依托硅基光子集成技术,在内部用光完成矩阵运算与数据交换。它的计算过程与人工智能算法高度匹配,计算速度比普通电子芯片高,功耗比电子芯片低。

    中科院上海微系统所所长助理、上海新微科技集团总裁秦曦曾介绍,集成电路的发展沿着摩尔定律已趋于极限,硅光子技术是超越摩尔研究领域的发展方向之一。通过硅光集成,用光代替原来的电进行传输,成本有可能降低到原来的十分之一,甚至更低。

    释疑2 与传统芯片比有哪些优势?

    白冰解释,主要有两方面优势。一个优势是计算速度,光子人工智能芯片的计算速度大概是电子芯片的三个数量级,约1000倍,单个电子芯片的计算速度大约是7.8TFlops,而光子人工智能芯片的计算速度大概是3200TFlops。第二个优势是功耗,光子人工智能芯片的功耗仅为电子芯片的百分之一,单位电子芯片和耗电量大概300W,对应的光子人工智能芯片的耗电量只有4W。

    对比不同芯片在同一情境下是否具有优势,要考虑性能功耗比、单位美元提供算力两方面。性能功耗比是指消耗单位瓦特提供的性能,重在强调涉及多少电费,单位美元提供算力则重在强调芯片的生产成本。在这两方面,光子人工智能芯片比电子芯片更有优势。

    释疑3 未来主要应用在哪些方面?

    图丨光子突触作用原理

    由 AI 所带起的计算需求不断膨胀,促进了处理架构持续推陈出新,譬如英特尔未来将结合 CPU 与 FPGA 计算能力,借以应对更复杂的应用情境;英伟达则是在其最新一代的 GPU 方案上大幅强化推理性能。除此之外,亦有不少希望能够针对特定计算推出更适合的新架构,比如说类神经网络芯片 (NPU)、量子计算机 (Quantum computing),以及最新的计算概念:基于光子回路 (Photonic Circuits) 的计算架构。

    实际上,“光”被使用在计算环境中已经有超过数十年的历史,过去主要用以在不同芯片或存储设备间传输数据之用,而因为相关的传输技术成本太高,且必须搭配昂贵的周边才能显现出其效益,也因此,“光”的传输从没有被普及到消费市场,导致我们对这个事实没有太明确的认知。

    然而,计算则是另一层次的问题。

    [​IMG]


    图丨SMART Photonics的光子芯片

    白冰表示,光子人工智能芯片可广泛用于手机、安防监控、自动驾驶、服务机器人、无人机、工业物联网、企业服务器和数据中心等关键人工智能领域。比如在分拣机器人机械臂上装上摄像头,让它识别有什么东西,控制它去抓取等。

    据其介绍,光子人工智能芯片的发展得益于人工智能的发展。光学计算芯片其实在实验室一直存在,但它一直没有比较好的应用场景,没有办法落地应用。近年来伴随着人工智能的兴起,人工智能的算法特征恰好跟光学芯片物理性能匹配,这使得光学计算有了走出实验室、走向产业应用的机会,就是这样一个过程。

    中科院“百人计划”专家、上海微系统所研究员余明斌回国后也一直致力于硅光工艺平台的建设。他认为,这是一个可能引发巨大产业新空间的前瞻性技术。

    光子芯片原理及世界发展现状

    从传输走向计算,光子芯片将成终极计算解决方案?

    从 2006 年开始,英特尔推出首款标准 CMOS 工艺的电子混合硅激光器之后,电和光这两个截然不同的物理现象终于成功被凑在一起。往后数年,基于此技术的超高带宽光学传输架构更成为高性能数据中心的最爱,借此有效降低了大量数据传输造成的系统瓶颈。

    2015 年,IBM 研究人员,发表了针对光子计算的新实验性技术,通过把硅光子数组集成到与 CPU 相同的封装尺寸中。硅光子技术的问题一直在于芯片的光学接口,不过 IBM 的光子解决方案能被应用于系统单芯片 (SoC),以廉价的标准连接器 (edge connector) 在芯片之间传输光,或是只要将 CMOS 芯片边缘接在一起就能进行芯片对芯片的通讯。

    这些光子芯片的发展主要是作为解决传统芯片与芯片之间,或芯片与存储系统之间的互连问题。而凭借集成度高的光子芯片的发明,取代了过去庞大复杂的光传输架构,且速度能更快,延迟更低。

    然而,真正把“光子”带往计算领域,甚至架构成“光子芯片”的概念,却是近两年才逐渐被发掘出来。

    由于半导体芯片技术虽依靠新应用与算法的整合,能做到的事情也越来越多,但实际上芯片架构本身还是基于同样的逻辑之下,且受限于半导体工艺,计算能力、规模以及功耗、成本形成难以均衡的四角关系。

    这时,业界也开始积极寻找能突破现况的新计算技术。GPGPU、神经网络芯片、DSP、FPGA 都在不同的时期被提出来,擅长解决特定应用计算领域,但这些芯片并没有解决根本的问题,也就是其基于半导体结构所面临的物理特性限制。


    [​IMG]

    很简单的概念解释光子计算芯片,就是在芯片上使用了无数个光学开关器,作用就类似半导体芯片中的逻辑栅,利用不同波长,相位和强度的光线组合,在复杂的反射镜、滤波器以及棱镜结构所组成的数组中进行信息处理。

    硅光子和微电子一样,都是基于硅材料的半导体架构。而硅作为光学通信传输方面的应用已经相当普及,由于光的快速反应和并行特性,能瞬间传输大量数据,因此被普遍应用在数据中心的服务器上。也因为光子传输过程稳定,并行能力强,且纠错设计相对简单,传输和转换所需要的能量极小,所以采用光子计算的架构理论上可以做到相对低的功耗表现。其次,光子芯片理论上也能做到规模极小的应用上,比如说移动设备中。

    光子芯片可沿用目前成熟的半导体工艺技术,而目前仍处于实验阶段的光子芯片仅需要老旧的微米级工艺就可达到大幅超越既有半导体芯片的计算能力,也因此未来工艺微缩空间极大。而凭借芯片密度的增加,性能还能大幅成长,甚至有机会彻底改写摩尔定律的限制。

    沿用 CMOS 工艺是光子计算最大优势,但目标非取代传统半导体计算

    由于光子芯片基本上还是以目前的 CMOS 制造工艺为基础,相对于量子计算使用的特殊工艺,在成本或量产技术方面都要更有优势,虽然目前实验室中的光子芯片在密度上还比不过传统半导体芯片,但已经比量子芯片好很多了。

    而光子芯片的效能取决于架构和算法,比如说同时使用多少路不同波长的光来进行组合,或者是在芯片中使用的光学信号的带宽,以及光电转换时的瓶颈,但是单从光的物理特性上来看,在合适的算法上要做到传统半导体芯片的百倍速度是不会有太大的问题的。

    当然,理论上光子芯片可以做到规模很大,也可以做到很小,但因为光不适合做非线性运算,另外光芯片的集成度和尺寸还是会有一定的规范,要完全取代半导体芯片还是有很大的难度。

    从芯片、算法到周边的生态正在发展中

    2017 年 6 月,麻省理工学院研究团队针对可程序设计纳米光子处理器提出了一份论文,并且发表在《自然-光子》杂志上。该论文的第一作者及通讯作者,出生在杭州的沈亦晨目前为 Lightelligence 的联合创始人兼 CEO。

    沈亦晨强调,目前 Lightelligence 的光子芯片发展已经完成实验室阶段的展示,在算法、总线以及存储方面都有相对应的设计正在进行,当然,计算芯片最重要的还是生态,这点也需要更多科研机构和公司加入到扩展光学计算这一领域来共同建立。

    因为主力产品是芯片,所以核心部分在于算法和硬件的结合,以及相对应的芯片指令以及编译程序,而 Lightelligence 的工作就是要让开发出来的芯片可以应用到目前市场上流行的框架中,比如说 TensorFlow、Caffe 等。

    [​IMG]

    另外,由于光子计算在传输或者是存储有其特殊性, Lightelligence 也在开发相对应的周边设计。当然,沿用目前的存储系统虽可加快落地商用速度,但可能就会限制光子计算的性能表现,因此这部分未来还是会以搭配针对光子计算优化的设计为目标,才更能凸显光子计算的整体优势。

    如今 Lightelligence 团队正努力改善光子计算的相关生态,目前当然还不成熟,不过业界对于高性能计算,甚至更好的神经网络计算架构有着非常高的期待,相信其光子计算架构落地之后,可以大大加速整体 AI 计算生态的变革。

    沈亦晨表示,不论是特定用途,或者是针对通用计算能力,这个都会是芯片架构发展的不同过程的选择。Lightelligence首先还是会以技术或应用场景比较成熟的光子芯片应用着手,然后再逐步去扩大可应用的范围。同时也在努力开发光子芯片前后端的技术,为未来不同的计算场景进行更好的适配。

    沈亦晨强调,总体来讲,在实现光子计算的路上还有很多重大的工程改进需要完成,但和过去的种种光子计算的尝试相比,现在可能是最好的时机,也是最接近实现的一次。

    光子计算在处理一些 AI 算法时有独特的优势

    事实上,光子芯片或许将会是未来最适合用来作为 AI 计算的硬件架构,这是因为光的特性先天适合线性计算(AI 计算里最重要的部分),这包含了高维度的并行计算。相对的,虽然量子计算近来也因为 AI 而受到关注,但量子计算还是比较偏向擅长解码或搜索的领域,另外在量产的生态上也还不太成熟,但潜力却不容小觑。
     
  8. Hendrik_2000
    Online

    Hendrik_2000 Brigadier

    Joined:
    Dec 20, 2006
    Messages:
    7,634
    Likes Received:
    26,359
    Semiconductor makers boosted by AI trend
    By He Wei in Shanghai | China Daily | Updated: 2019-01-11 10:17

    [​IMG]
    Technicians check chips at a technology company in Guigang, the Guangxi Zhuang autonomous region.[Photo by Tan Kaixing/for China Daily]

    The increasing commercialization of artificial intelligence is set to give China a substantial boost in the global semiconductors field, say international researchers.

    Revenues from semiconductors manufactured in China will grow by 25 percent to approximately $110 billion in 2019, as producers meet the increasing domestic demand for chipsets, fueled in part by AI advances, said consultancy Deloitte Global in an annual industry preview released on Tuesday.

    A Chinese chip foundry will begin producing semiconductors specialized to support AI and machine learning tasks, thanks to massive domestic demand and the technological might of domestic tech giants, said Chris Arkenberg, a research manager with Deloitte's Center for Technology, Media and Telecommunications.

    "China is perhaps better positioned now than ever before to become a globally competitive player in both semiconductors and AI ... because leading digital businesses (in China) have signaled that greater domestic self-supply of semiconductors is a vital component of their future," he said.

    For instance, China's technology trio, namely Baidu, Alibaba and Tencent, hold stakes in more than half of China's 124 unicorn startups, including SenseTime, the world's most valuable pure-play AI company.

    "They are spending and hiring aggressively to create onshore manufacturing capabilities approaching those of the top global foundries," Arkenberg noted.

    Beijing's Horizon Robotics, founded by the former head of Baidu's Institute of Deep Learning, supplies embedded chips for machine vision and is working with major automotive brands to provide edge processing with machine vision for vehicles.

    Another notable Chinese chip player, Cambricon, also has a line of chips specializing in supporting machine learning tasks, contributing design support for AI in Huawei's Kirin smartphone chipset and then delivering its own machine learning solutions for data centers.

    The massive troves of data in the Chinese market will help to improve the precision and accuracy of algorithms, thus fueling the development of AI chips, said Roger Chung, Deloitte Research TMT senior manager.

    In announcing the decision to establish a dedicated chip company in October, Zhang Jianfeng, chief technology officer of Alibaba, attributed the tech giant's unique position to lead breakthroughs in chips to its "advantages in algorithm and data intelligence".

    AI is likely to become a springboard for China's semiconductors industry in the long term, given the massive troves of data generated in various scenarios and the relatively easier access to them, said Bill Lu, a Hong Kong-based managing director in research at UBS.

    "China's top internet players have the biggest command of consumer data and are better positioned to bear the ever-increasing marginal costs to capture new customers," Lu said.

    Advances in AI are one of the driving forces for the global semiconductor industry with an anticipated 5 to 6 percent growth rate over the next two decades, said Morris Chang, founder of Taiwan Semiconductor Manufacturing Co, in September.
     
  9. Hendrik_2000
    Online

    Hendrik_2000 Brigadier

    Joined:
    Dec 20, 2006
    Messages:
    7,634
    Likes Received:
    26,359
    Via QWERTY

    more detail on new robosense solid-state lidar
    -----



    prnewswire.com
    CES 2019 Innovation Award Honoree RoboSense Launches New MEMS Solid-State LiDAR at CES 2019
    RoboSense
    6-7 minutes
    SHENZHEN, China, Jan. 3, 2019 /PRNewswire/ -- RoboSense, a leader in LiDAR perception technology solutions and CES 2019 Innovation Award Honoree, announced today that next week at CES 2019, they will publicly demonstrate an upgraded version of their MEMS solid-state LiDAR, an automotive grade solid-state LiDAR designed for the mass production of autonomous vehicles. The new RS-LiDAR-M1 with patented MEMS technology offers ground-breaking vehicle intelligence awareness to fully support Level 5 driverless automated driving. A breakthrough on the measurement range limit based on 905nm LiDAR with a detection distance to 200 meters, the upgraded optical system and signal processing technology brings remarkable final output point cloud effect which can now clearly recognize even small objects, such as railings and fences.

    At last year's CES 2018, RoboSense demonstrated the first generation MEMS solid-state LiDAR RS-LiDAR-M1Pre. Just four months later, in May 2018, it was loaded on the Cainiao unmanned logistics vehicle and unveiled at the Ali CainiaoGlobal Intelligent Logistics Conference, becoming the world's first solid-state LiDAR for unmanned vehicles. RoboSense has already been sending the MEMS LiDAR product to the world's top OEMs and Tier 1 suppliers. At CES 2019, RoboSense will launch the new RS-LiDAR-M1, showcasing the potential of their MEMS optomechanical system design, with breakthrough improvements in detection distance, resolution, Field of View (FOV), reliability, and other RoboSense award-winning LiDAR sensing technologies.

    Double the Horizontal Field of View to 120 Degrees

    A major step forward from the previous version award-winning RoboSense RS-LiDAR-M1Pre, the new RS-LiDAR-M1 MEMS optomechanical LiDAR provides an increased horizontal field of view by nearly 100% compared to the previous generation, reaching an amazing 120° field of view, so that only a few RS-LiDAR-M1s are needed to cover the 360° field of view. In addition, with only five RS-LiDAR-M1s, there is no blind zone around the carwith dual LiDAR sensing redundancy provided in front of the car for a L5 level of automatic driving -- full driverless driving. Based on the target production cost at $200 each, the cost of five RS-LiDAR-M1 is only 1/100th the highest mechanical LiDAR available to the market, which is more inline with the cost requirements for the mass production of autonomous vehicles.

    Measurement Range to 200 Meters

    The battle between 1550nm and 905nm LiDAR is about cost and performance. When aiming for a low-cost 905nm LiDAR, it is necessary to overcome the technical difficulties of achieving sufficient measurement range. The RS-LiDAR-M1 achieves a breakthrough on the measurement range limit based on the 905nm LiDAR, with a detection distance to 200 meters.

    A Leap Forward in Point Cloud Effect Technology -- Even Small Objects

    The unique RS-LiDAR-M1 LiDAR system provides massive improvements, the most remarkable being the final output point cloud effect. The M1's detection capability is greatly improved through the upgraded optical system and signal processing technology, which can now clearly recognize even small objects,such as railings and fences.

    "The RoboSense RS-LiDAR-M1 LiDAR system is a giant leap forward for driverless technology," said Mark Qiu, Co-founder, RoboSense. "We are committed to developing high-performance automotive-grade LiDAR at a low-cost to advance the LiDAR market, so that LiDAR can be used in fully unmanned vehicles, as well as assisted autonomous driving with superior environmental information detection that ensures driving safety."

    A CES 2019 Innovation Award Honoree, the RS-IPLS first hardware and software algorithm based solution for the mass production of safer autonomous cars includes the RS-LiDAR-M1Pre first generation RoboSense MEMS LiDAR system and AI algorithms. RoboSense will be exhibiting the new upgraded MEMS LiDAR, the RS-LiDAR-M1, at CES 2019, Booth #9310, North Hall. CES is held January 8-11, 2019 at the Las Vegas Convention Center in Las Vegas, NV.
     
  10. Hendrik_2000
    Online

    Hendrik_2000 Brigadier

    Joined:
    Dec 20, 2006
    Messages:
    7,634
    Likes Received:
    26,359
    Contrary to popular believe Fujian Jinhua is not dead YET via XYZ
    光刻机龙头ASML默默支持,福建晋华低调运转不停工
    https://xw.qq.com/cmsid/20190117A0GX3D
    国内 DRAM 技术阵营福建晋华在被美方列入出口管制清单当日,所有美系设备商全数撤出,之后甚至传出整个工厂陷入停摆,然事实上,福建晋华仍是低调持续运转,且最大光刻机龙头 ASML 默默支持并未撤出,目前由福建晋华总经理陈正坤坐镇,采“龟息大法”维持工厂正常运作。

    国内 DRAM 技术阵营福建晋华在被美方列入出口管制清单当日,所有美系设备商全数撤出,之后甚至传出整个工厂陷入停摆,然事实上,福建晋华仍是低调持续运转,且最大光刻机龙头 ASML 默默支持并未撤出,目前由福建晋华总经理陈正坤坐镇,采“龟息大法”维持工厂正常运作。


    (来源:ASML)

    2018 年 10 月 29 日美国一纸禁令宣布晋华被列为出口管制清单中,隔天 30 日,美系设备、材料等供应商全面撤出,包括机台安装、协助生产等动作也立刻停止,之后传出连非美系的供应商也撤出,整个 12 寸厂房呈现停摆状态。

    事实上,目前福建晋华的 12 寸厂房仍是低调运转中,且光刻机大厂 ASML 并未如外传撤出,仍是默默支持,ASML 对 DT 君表示,会在符合各种法规要求的前提下,正常支持客户。

    很显然地,虽然遭遇巨大冲击和压力,晋华也走到此一险境,但并没有放弃 DRAM 技术计划,现阶段的晋华员工仍是抱着没有退路的信念继续做下去,战到最后一刻。

    受到晋华事件冲击,国内另一个 DRAM 技术阵营合肥长鑫的动态则是更为低调。

    业界透露,合肥长鑫的 DRAM 芯片已经研发出来,良率也不错,但碍于眼前的“晋华风暴”,可能会延后量产。

    再者,由于之前不少美光前员工加入合肥长鑫的团队,为了规避目前锋头正盛的 DRAM 专利侵权疑虑,长鑫也雇用不少来自前三星电子(Samsung Electronics)、 SK 海力士(SK Hynix)的韩国研发团队,竭尽所能地淡化美系色彩。

    不过,传出之前“晋华案”爆发后,这些韩籍员工返回韩国后,有被当地政府“关切”,显示国内自制存储“芯”的风潮,仍是处于国际间的严密注视中。

    眼前的福建晋华几乎是采用“龟息大法”在维持生命状态,让工厂的运作继续撑下去,但来自三面的压力仍是持续。一方面是来自美国司法部的起诉已在美国旧金山联邦法院开庭,晋华、联电提出无罪抗辩,争取从美国商务部出口管制清单中移除。

    面对美国司法部的起诉,晋华和联电在美国分别寻找适合的律师协助处理该案,然而,晋华也遇到不少阻碍,主要是因为美国多家著名的律师事务所都与美系存储大厂美光(Micron)签约,而美光正是“晋华案”的幕后主要推手。

    后来,晋华找到的律师为曾担任联邦检察官的 Christine Y. Wong,联电则是找来之前在美国欧巴马时代,担任联邦司法部助理总检察长的 Leslie R. Caldwell,作为辩护律师。

    晋华遇到第二层面的压力,则是技术合作伙伴联电的终止 DRAM 技术开发计划。

    根据原本规划,联电接受晋华委托的 DRAM 技术开发计画是在联电南科厂进行研发,待研发成功后,该组百人的 DRAM 研发团队将从联电转移至晋华,只是这组人还没转过去,美国的出口禁令就已经先一步出手拦截,现在这一组百人团队在联电已经内转到其他部门。

    除了上述的美国司法大刀、合作伙伴退出 DRAM 技术开发计画外,晋华遇到的第三层阻碍是来自美系供应商的撤出。

    2018 年 10 月底美国出口管制禁令一出,所有美系供应商包括 Applied Materials 、 KLA Corporation 、 Lam Research 、 Gatan 、 ECI Technology 、 Broker 、 Engegris 等全数撤出晋华,不过,部分非美系供应商并没有撤出,且晋华的晶圆厂仍是维持正常运作。

    由于福建晋华先前已经采购可生产几千片的半导体机台设备,用以支应今年初进入小量投产,现在这些设备虽没有美系设备商人员协助生产,但晋华仍是想办法让机台正常运作,在耗材原物料方面,则是找韩系、日系的供应商代替美系的料源。

    再者,合肥长鑫的 DRAM 技术开发计划是基于北京兆易创新、合肥市产业投资控股签署的《关于存储器研发项目之合作协议》,进行 19nm 工艺研发,总预算约为 180 亿元人民币,双方筹资比例为 1 :4,目标是在 2018 年 12 月 31 日前实现产品良率 (测试电性良好的晶片占整个晶圆的比例) 不低于 10% 目标。

    兆易创新也于 2018 年 12 月 28 日公告,经沟通确认,双方约定合作的 DRAM 研发项目不会暂停

    The lithography machine faucet ASML silently supports, Fujian Jinhua low-key operation does not stop https://xw.qq.com/cmsid/20190117A0GX3D domestic DRAM technology camp Fujian Jinhua on the day of the US listed on the export control list, all US equipment vendors After the withdrawal, all the factories even went into a lockout. In fact, Fujian Jinhua is still in a low-key operation, and the largest lithography machine leader ASML silent support has not been withdrawn. Currently, Chen Zhengkun, general manager of Fujian Jinhua, is sitting. Adopting the "great method" to maintain the normal operation of the factory. On the day when the domestic DRAM technology camp Fujian Jinhua was listed on the export control list by the US, all the US equipment vendors were withdrawn, and even the entire factory was shut down. In fact, Fujian Jinhua is still low-key and continuous operation, and the largest The silent support of the lithography machine faucet ASML has not been withdrawn. At present, Chen Zhengkun, the general manager of Fujian Jinhua, is sitting in the town and adopts the “great interest law” to maintain the normal operation of the factory. (Source: ASML) On October 29, 2018, a US ban announced that Jinhua was listed as an export control list. On the next day, the US equipment, materials and other suppliers were fully withdrawn, including machine installation and assistance in production. The other actions were immediately stopped, and then the non-US suppliers were also withdrawn, and the entire 12-inch factory was shut down. In fact, the current 12-inch factory in Fujian Jinhua is still in low-key operation, and the lithography machine manufacturer ASML has not been withdrawn from the outside, and it is still silently supported. ASML said to DT Jun that it will meet various regulatory requirements. Under the premise, the customer is normally supported. Obviously, despite the tremendous impact and pressure, Jinhua has also come to this danger, but has not given up the DRAM technology plan. At this stage, Jinhua employees are still carrying on the belief that there is no retreat, and they will continue to do so. A moment. Affected by the Jinhua incident, the dynamic of another domestic DRAM technology camp, Hefei Changxin, is even more low-key. According to the industry, Hefei Changxin's DRAM chip has been developed, and the yield is also good. However, due to the “Jinhua Storm”, it may delay production. In addition, since many former Micron employees joined the team of Hefei Changxin, Changxin also hired many former Samsung Electronics and SK Hynix (SK) in order to avoid the current DRAM patent infringement concerns. Hynix's Korean R&D team is doing everything possible to downplay American colors. However, after the outbreak of the Jinhua case, these Korean employees returned to South Korea and were “concerned” by the local government, indicating that the domestic self-made storage “core” trend is still in the international close watch. In front of Fujian Jinhua, almost using the "great method" to maintain the state of life, let the operation of the factory continue, but the pressure from the three sides is still continuing. On the one hand, the lawsuit from the US Department of Justice has been held in the Federal Court of San Francisco, and Jinhua and UMC have pleaded not guilty and sought to remove it from the US Department of Commerce's export control list. In the face of the US Department of Justice's lawsuit, Jinhua and UMC searched for suitable lawyers in the United States to help deal with the case. However, Jinhua also encountered many obstacles, mainly because many famous law firms in the United States and the United States The storage giant Micron signed the contract, and Micron is the main promoter behind the Jinhua case. Later, Jinhua found the lawyer who was a federal prosecutor, Christine Y. Wong, and the company was Leslie R. Caldwell, who served as assistant attorney general of the Federal Ministry of Justice in the US Obama era, as a defense lawyer. Jinhua's second-level pressure is the termination of DRAM technology development plan by UMC. According to the original plan, UMC accepted the DRAM technology development plan commissioned by Jinhua to conduct research and development at UMC Nanke Factory. After the successful development, the group's DRAM R&D team will be transferred from UMC to Jinhua, but this group has not yet In the past, the US export ban has been intercepted first, and now this group of 100 people has been transferred to other departments in UMC. In addition to the above-mentioned US judicial slashers and partners withdrawing from the DRAM technology development program, the third obstacle encountered by Jinhua is the withdrawal from US suppliers. At the end of October 2018, the US export control ban came out. All US suppliers including Applied Materials, KLA Corporation, Lam Research, Gatan, ECI Technology, Broker, Engegris, etc. all withdrew from Jinhua, but some non-US suppliers There was no withdrawal and Jinhua’s fab was still operating normally. Since Fujian Jinhua has previously purchased thousands of semiconductor machine equipment, it will be used to support a small amount of production at the beginning of this year. Now, although these equipments are not assisted by American equipment manufacturers, Jinhua is still trying to make a machine. The station is in normal operation. In terms of raw materials for consumables, it is to find Korean and Japanese suppliers to replace the US source. In addition, Hefei Changxin's DRAM technology development plan is based on Beijing Zhaoyi Innovation and Hefei Industrial Investment Holdings' “Cooperation Agreement on Memory R&D Project”, which will carry out 19nm process research and development with a total budget of approximately RMB 18 billion. The fundraising ratio is 1:4, and the goal is to achieve a product yield (the proportion of wafers with good electrical conductivity to the entire wafer) of no less than 10% by December 31, 2018. Zhaoyi Innovation was also announced on December 28, 2018. After confirmation by communication, the DRAM R&D projects agreed by the two parties will not be suspended.
     
Loading...

Share This Page