景況不佳的日本半導(dǎo)體廠商們終于覺醒并開始面對一個現(xiàn)實(shí)──擁抱“輕晶圓廠(fab-lite)“策略將會是他們存活的最佳希望；但很大的問題是：誰會想在日本半導(dǎo)體產(chǎn)業(yè)的“清倉大拍賣”中收購一座舊晶圓廠？ 答案恐怕是沒有人。市場研究機(jī)構(gòu) Future Horizons 董事長暨首席執(zhí)行官 Malcolm Penn 話就說得很直，他表示那些日本晶圓廠“都是采用舊制程”，而且誰要是買下那些晶圓廠：“就會面臨管理以及物流(logistic)的夢靨?！? 不過換個角度看，因?yàn)槟愕馁€注是押在沒人看好的那一方，意味著可能贏得的彩金將加倍。 最近日本東京街頭巷尾流傳著一個八卦，說有海外公司打算出價收購某日本芯片業(yè)者試圖放棄的晶圓廠。不過大多數(shù)產(chǎn)業(yè)界人士都無法理解，真會有哪家公司膽子這么大──或是這么笨──會去做這樣的事情。 Penn表示：“如果它們已不再優(yōu)秀、對研發(fā)沒有用處，為何要收購它們？由一個純教科書的市場經(jīng)濟(jì)觀點(diǎn)來看，讓它們下線并消失是比較好的做法?！钡攸c(diǎn)來了：要關(guān)閉一座晶圓廠的成本不小，包括得花錢雇人把廠房拆毀。 有鑒于此，讓一家海外芯片業(yè)者用折扣價買下那個龐然巨物并讓它能繼續(xù)運(yùn)轉(zhuǎn)，并不是那么牽強(qiáng)的想法。站在“某日本芯片廠商”的觀點(diǎn)，邀請一家海外同業(yè)進(jìn)來用他們的晶圓廠制造產(chǎn)品，同時滿足其現(xiàn)有客戶的需求，會比承受兩到三倍的虧損關(guān)閉整座廠房來得好。 另一家市場研究機(jī)構(gòu)Semico Research的主管Joanne Itow表示，舊晶圓廠的價格根據(jù)其廠房規(guī)模、技術(shù)種類、所制造的產(chǎn)品不同，會有很大的差異──但如果關(guān)閉一座晶圓廠的成本大約要3億美元，為何不用2億美元甚至1億美元的價格賣掉它？ Itow認(rèn)為收購日本的舊晶圓廠是有潛在商機(jī)的，事實(shí)上已經(jīng)有一個先例：“Tower收購過美光(Micron)旗下位于日本的晶圓廠?！倍徽J(rèn)為一線晶圓代工廠會有興趣在日本經(jīng)營晶圓廠，指一些專注于MEMS、電源管理等特定技術(shù)的純晶圓代工廠，較有可能成為買主。 此外Itow指出，如果不是買下整座晶圓廠，一線晶圓代工業(yè)者可能會有興趣收購那些日本晶圓廠內(nèi)的二手設(shè)備，例如格羅方德(GlobalFoundries)最近收購臺灣茂德科技(ProMOS)的晶圓廠設(shè)備；她認(rèn)為這是可能發(fā)生的：“隨著臺積電(TSMC)打算擴(kuò)充CIS、HV與eFlash產(chǎn)能，他們可能會涉足二手設(shè)備市場?！? 以上Itow所舉的例子，都是關(guān)于純晶圓代工業(yè)者可能對日本舊晶圓廠產(chǎn)生潛在興趣；那么來自美國的無晶圓廠芯片業(yè)者呢？會有人想買嗎？ 本文授權(quán)編譯自EE Times，版權(quán)所有，謝絕轉(zhuǎn)載 第2頁：MEMS廠商可能會由無晶圓廠經(jīng)營型態(tài)，改采輕晶圓廠策略 第3頁：日本半導(dǎo)體業(yè)者旗下晶圓廠概況

相關(guān)閱讀：
• 被英特爾搶代工訂單，臺積電加速與富士通同盟抗美韓
• GlobalFoundries坐等瑞薩裁掉的專業(yè)人才
• “輕晶圓廠”風(fēng)行，IDM加快IC制造外包3YDesmc

{pagination} 在眾多芯片廠商一面倒地選擇無晶圓廠或輕晶圓廠經(jīng)營模式的此刻，買座晶圓廠的想法幾乎是不可能──除非這樣的交易對買賣雙方來說都是具吸引力的；這不僅需要超凡的想象力，對買方來說也需要很大的信心勇氣。 有 少數(shù)美國硅谷的半導(dǎo)體廠偶爾會說，無晶圓廠芯片業(yè)者需要擁有取得晶圓廠產(chǎn)能的暢通管道，特別是當(dāng)這家公司正在開發(fā)新組件構(gòu)造或結(jié)構(gòu)，若無法掌控制程技術(shù)， 就無法進(jìn)行概念的實(shí)驗(yàn)，更別說去理解如何在開發(fā)過程中打造新架構(gòu)；而且更重要的是，晶圓代工業(yè)者通常都不愿意跟那些還沒簽下大客戶的無晶圓廠芯片業(yè)者合 作。 因此誰能幫忙那些無晶圓廠芯片業(yè)者測試新概念？在硅谷，有像是SVTC Technologies那樣的技術(shù)服務(wù)公司，可提供半導(dǎo)體產(chǎn)品開發(fā)與商業(yè)化的制程技術(shù)；該公司在美國加州圣荷西、德州奧斯汀等地有晶圓廠，但在去年縮減業(yè)務(wù)。 對此Itow表示：“我懷疑小型無晶圓廠業(yè)者是否會收購在日本的晶圓廠，因?yàn)橘Y本支出負(fù)擔(dān)不小?！彼⒅赋觯瑹o晶圓廠業(yè)者：“還有不少可供產(chǎn)品研發(fā)與新組件架構(gòu)測試的其它制程服務(wù)選擇，包括加州柏克萊大學(xué)、亞利桑那州立大學(xué)等都可提供原型制造服務(wù)?！? Future Horizon的Penn表示，歐洲研發(fā)機(jī)構(gòu)IMEC也提供這類服務(wù)；不過Itow反駁該說法，指IMEC不能被看作是制造晶圓廠：“但他們確實(shí)能做為提 供制程開發(fā)與產(chǎn)品研發(fā)服務(wù)的選項(xiàng)?！彼硎?，可提供制程服務(wù)的，還有美國阿爾巴尼大學(xué)納米科學(xué)與工程學(xué)院(CNSE)。 在純晶圓代工業(yè)者以外，MEMS制造商也可能對日本舊晶圓廠有興趣；Itow表示：“MEMS制造商開始采用較大尺寸晶圓，正由6寸轉(zhuǎn)移至8寸?！倍患襇EMS供貨商Kionix就在日本擁有一座自己的晶圓廠。 Itow 指出，MEMS廠商可能會由無晶圓廠經(jīng)營型態(tài)，改采輕晶圓廠策略；Alpha Omega Semiconductor就是一例，這家公司專長離散/功率芯片，在去年主要是采用華虹NEC (HHNEC)的制程：“2011年，Alpha Omega收購了IDT位于美國奧勒岡州、一座已16歲的晶圓廠，被列為一家輕晶圓廠業(yè)者?！? 本文授權(quán)編譯自EE Times，版權(quán)所有，謝絕轉(zhuǎn)載 第3頁：日本半導(dǎo)體業(yè)者旗下晶圓廠概況

相關(guān)閱讀：
• 被英特爾搶代工訂單，臺積電加速與富士通同盟抗美韓
• GlobalFoundries坐等瑞薩裁掉的專業(yè)人才
• “輕晶圓廠”風(fēng)行，IDM加快IC制造外包3YDesmc

{pagination} 日本半導(dǎo)體業(yè)者旗下晶圓廠概況 富士通半 導(dǎo)體(Fujitsu Semiconductor)已經(jīng)在去年秋天將旗下位于日本巖手縣(Iwate)的廠房移交給電裝株式會社(Denso Corp.)；富士通位于日本會津(Aizu)、宮城縣(Miyagi)與九州(Kyushu)的封裝生產(chǎn)線，則轉(zhuǎn)手給J-Devices。 去年12月，富士通位于日本三重縣(Mie)的12寸晶圓廠仍乏人問津，該公司表示，他們目前正：“考慮將之轉(zhuǎn)賣給晶圓代工廠，包括臺積電。”

瑞薩旗下晶圓廠分布圖(2011年)，黃色圓點(diǎn)所標(biāo)示的是前段制程晶圓廠，紫色三角形標(biāo)示的為后段封測廠
Source：Renesas3YDesmc

在此同時，瑞薩(Renesas)雖然已經(jīng)宣布進(jìn)行組織重整，該公司半導(dǎo)體前段制造部門還有7個生產(chǎn)據(jù)點(diǎn)、9條生產(chǎn)線；后段制造部門則有2個生產(chǎn)據(jù)點(diǎn)。 瑞薩在日本青森縣的后段制造據(jù)點(diǎn)Renesas High Components，已經(jīng)在1月轉(zhuǎn)手Aoi Electronics；另三座分別位于函館(Hakodate)、福井(Fukui)與熊本(Kumamoto)的后段制造據(jù)點(diǎn)，則將在6月初賣給J- Devices。 雖然瑞薩尚未表明那些剩余的生產(chǎn)據(jù)點(diǎn)將會轉(zhuǎn)賣、關(guān)閉或是縮減產(chǎn)能規(guī)模，在今年稍早，該公司表示計(jì)劃在截止于2014年3月的財(cái)務(wù)年度，將折舊成本與租賃費(fèi)用(depreciation cost and lease fees)比上一財(cái)務(wù)年度縮減25%。 本文授權(quán)編譯自EE Times，版權(quán)所有，謝絕轉(zhuǎn)載 編譯：Judith Cheng 參考英文原文：Japanese fabs, anyone?，by Junko Yoshida

相關(guān)閱讀：
• 被英特爾搶代工訂單，臺積電加速與富士通同盟抗美韓
• GlobalFoundries坐等瑞薩裁掉的專業(yè)人才
• “輕晶圓廠”風(fēng)行，IDM加快IC制造外包3YDesmc

{pagination} Japanese fabs, anyone? Junko Yoshida The latest chatter on the Tokyo street reveals that at least one foreign company is bidding for a fab that a Japanese chip company is trying to dump. TOKYO--As ailing Japanese semiconductor manufacturers begin finally to awaken to a reality that going fab-lite is their best hope for survival, one question looms large: Who on earth would want to buy an old fab at a Japanese rummage sale? The conventional wisdom says nobody. Malcolm Penn, chairman & CEO at Future Horizons, flatly said that those Japanese fabs “are all in legacy nodes,” and buying any of those fabs “would be a management and logistic nightmare.” On the other hand, when you bet on an opportunity that appeals to nobody else, the potential jackpot is all that much bigger. The latest chatter on the Tokyo street reveals that at least one foreign company is bidding for a fab that a Japanese chip company is trying to dump. Most industry insiders, however, can’t fathom why any company would be bold--or foolish--enough to make such a move. Penn said, “If they’re not good enough anymore to be useful to R&D, why should I buy them? Better from a pure textbook market economic point of view to take them off line and raze them.” But here’s the thing: It costs a lot to close a fab, with expenses covering everything from paying off employees to gutting the site. In this light, the idea of letting a foreign chip company buy the white elephant at a discount and having them run it isn’t so far-fetched. From the viewpoint of the Japanese chip company, inviting a foreign company in, to use the fab for its products while meeting the demands of existing customers, is a better outcome than shutting the whole thing down at two or three times the loss. The pricing of old fabs differs significantly depending on the size of the fab, type of technology, products produced, said Joanne Itow, managing director at Semico Research. But just for the sake of argument, what if it could cost as much as $300 million to shut down a fab? In that case, why not sell for $200 million--or even $100 million? Itow is an analyst who sees potential opportunity in the acquisition of old fabs in Japan. Indeed, there is a precedent. She said, “One example is Tower’s purchase of a fab located in Japan from Micron.” While Itow doesn’t believe that Tier 1 foundries are interested in operating a fab in Japan, she pointed out that there are several pure-play foundries focused on MEMS, power management, and other products. They could become buyers. Even if it were not an outright purchase of the whole fab, Itow thinks there could be some interest from Tier 1 foundries to purchase equipment already installed in the Japanese fabs. Pointing out GlobalFoundries’ recent purchase of equipment from the ProMOS Taiwan fab, she believes this could be done. Further, “As Taiwan Semiconductor Manufacturing Co. expands its CIS, HV and eFlash capacity, they could be in the market for used equipment.” The examples given by Itow above are mostly about potential interest in old Japanese fabs among pure-play foundries. How about fabless chip companies in the United States? Any takers? Desire to own a fab At a time when so many chip companies are overwhelmingly opting for the fabless or fab-lite model, such a move is simply unimaginable--unless the deal is structured in a way that’s attractive to both buyer and seller. This may take not only genuine imagination but also a leap of faith on the buyer’s part. But a few vendors in Silicon Valley argue that sometimes a fabless chip company needs good access to a fab, especially if the company is developing new configurations or structures for circuit devices. Without access to process technologies, such a company couldn’t experiment with its ideas, let alone understand how to model new structures in development. More important, foundries are reluctant to work with fabless chip vendors who haven’t already signed up big customers. So who is there to help them test their new ideas? Silicon Valley used to have SVTC Technologies, a technology services company that provides development and commercialization services for semiconductor process-based technologies and products. It had facilities in San Jose, California and Austin, Texas. But SVTC cut back its operations last year. Itow, however, remains unconvinced. “I doubt if a small fabless company would buy a fab in Japan, because of the capex burden.” Further, she pointed out that fabless companies “have access to a number of other options for R&D, and testing of new device structures. Many universities, including the University of California-Berkeley, Arizona State Univ. and others, offer prototyping services.” Future Horizon’s Penn suggests IMEC to serve such needs. Itow countered that IMEC isn’t exactly viewed as a production fab. “But they would certainly be an option for process development and R&D,” she said. Other options, suggested by Itow, include the College of Nanoscale Science and Engineering at (CNSE)’s Albany NanoTech Complex. MEMS manufacturers? If not pure-play foundries, MEMS manufacturers could be among the candidates looking at old fabs in Japan. Itow said, “MEMS manufactures are beginning to take advantage of larger wafer sizes, i.e. moving from 6” wafers to 8” wafers.” One good example is Kionix who is a MEMS company and already owns a fab in Japan. MEMS companies could move from fabless to fab-lite, Itow pointed out. Alpha Omega Semiconductor is an example that changed strategy from fabless to fab-lite. The company specializes in discrete/power IC products and until last year primarily used HHNEC. “In 2011, Alpha Omega purchased a 16-year-old fab in Oregon from IDT. They are now classified as a fab-lite company,” she added. As for Fujitsu Semiconductor, the company completed the transfer of its Iwate Plant to Denso Corp. last fall. Fujitsu’s assembly lines in Aizu, Miyagi and Kyushu were transferred to J-Devices Corp. last December. Fujitsu’s 300mm wafer lines in Mie remain unsold, currently in “deliberation on transfer to a new foundry company, including TSMC,” according to Fujitsu. Renesas' fab locations as of 2011 Yellow circle: front-end wafer fabs Purple triangle: backend assembly test plants Source: Renesas Meanwhile, Renesas, even after announcing production structural reforms, will still have seven sites and nine lines left for its front-end production; and two sites for back-end. Renesas High Components, a back-end production site in Aomori, was transferred to Aoi Electronics Co., Ltd. in January. Three other back-end process sites – including Hakodate, Fukui and Kmamoto – are being sold to J-Devices in early June. While Renesas has not spelled out which sites – among those left – will be targets for the “transfer, closure and streamlining” of its production scales, the Japanese company said earlier this year that it plans to reduce depreciation cost and lease fees for the fiscal year ending March 2014 by approximately 25% from the fiscal year ending March 2013 and approximately 35% from the fiscal year ending March 2011.