泰瑞達 (TER) 2021 Q3 法說會逐字稿

Ladies and gentlemen, thank you for standing by, and welcome to the Teradyne Third Quarter 2021 Earnings Conference Call. (Operator Instructions) I would now like to hand the conference over to your speaker today, Andrew Blanchard.

各位女士、先生，感謝您稍候，歡迎參加 Teradyne 2021 年第三季財報電話會議。(接線員指示) 現在我想將會議交給今天的講者 Andrew Blanchard。

Thank you. Please go ahead, sir.

謝謝。先生，請開始。

Thank you, Patrice.

謝謝你，Patrice。

Good morning, everyone, and welcome to our discussion of Teradyne's most recent financial results. I'm joined this morning by our CEO, Mark Jagiela; and our CFO, Sanjay Mehta. Following our opening remarks, we'll provide details of our performance for 2021's third quarter along with our outlook for the fourth quarter.

各位早安，歡迎參加我們對 Teradyne 最新財務結果的討論。今天早上與我一同出席的有我們的執行長 Mark Jagiela，以及財務長 Sanjay Mehta。在開場致詞後，我們將提供 2021 年第三季的業績細節，並分享對第四季的展望。

The press release containing our third quarter results was issued last evening. We're providing slides on the Investor page of the website that may be helpful to you in following the discussion. Replays of this call will be available via the same page after the call ends.

包含我們第三季結果的新聞稿已於昨晚發布。我們在網站的投資人頁面提供簡報投影片，可能有助於您跟進本次討論。本次電話會議結束後，亦可透過同一頁面收聽重播。

The matters that we discuss today will include forward-looking statements that involve risk factors that could cause Teradyne's results to differ materially from management's current expectations. We encourage you to review the safe harbor statement contained in the earnings release as well as our most recent SEC filings. Additionally, those forward-looking statements are made as of today and we take no obligation to update them as a result of developments occurring after this call.

我們今天討論的事項將包含前瞻性陳述，其中涉及可能導致 Teradyne 的結果與管理層目前預期出現重大差異的風險因素。我們建議您查閱財報新聞稿中的安全港聲明，以及我們最新的 SEC 申報文件。此外，這些前瞻性陳述係以今日為準，我們不承擔因本次電話會議後發生之發展而更新其內容的義務。

During today's call, we'll make reference to non-GAAP financial measures. We've posted additional information concerning these non-GAAP financial measures where appropriate on the Investor page of the website. Looking ahead between now and our next earnings call, Teradyne expects to participate in technology or industrial-focused investor conferences hosted by Baird, Credit Suisse, Davidson and UBS.

在今天的電話會議中，我們將提及非 GAAP 財務衡量指標。我們已在網站投資人頁面（適用處）張貼與這些非 GAAP 財務衡量指標相關的補充資訊。展望從現在到下一次財報電話會議之間，Teradyne 預期將參與由 Baird、Credit Suisse、Davidson 與 UBS 主辦、聚焦科技或工業領域的投資人研討會。

Now let's get on with the rest of the agenda. First, Mark will comment on our recent results, current market conditions and thoughts on the rest of 2021 and '22. Sanjay will then offer more details on our quarterly results along with our guidance for the fourth quarter. We'll then answer your questions, and this call is scheduled for 1 hour.

現在讓我們進入其餘議程。首先，Mark 將評論我們近期的結果、目前的市場狀況，以及對 2021 年與 2022 年的看法。接著 Sanjay 將提供更多季度結果細節，並給出第四季的財測指引。之後我們將回答各位的問題，本次電話會議預計為 1 小時。

Mark?

Mark？

Thanks, Andy.

謝謝，Andy。

Good morning, everyone, and thanks for joining us. Today, I'll cover 4 topics: the highlights of our third quarter and the first 9 months of the year, the changes we're observing in the SOC test market, our outlook for the industrial automation market and how we're thinking about the test and automation markets as we close out this year and look into 2022 and beyond.

各位早安，感謝各位加入我們。今天我將涵蓋 4 個主題：第三季與今年前 9 個月的重點、我們在 SOC 測試市場觀察到的變化、我們對工業自動化市場的展望，以及在今年收官並展望 2022 年及更長期時，我們如何看待測試與自動化市場。

As our Q3 results demonstrate, demand remains strong across all of our businesses. At the company level, Q3 sales grew 16% from last year's record Q3 and non-GAAP EPS grew 35%. We did experience increased supply chain bottlenecks in our Industrial Automation business in the quarter and under ship demand. Sanjay will describe this in more detail, but we expect these constraints to persist into Q4. Despite this, for the first 9 months of 2021, company-wide sales grew 19% and non-GAAP EPS grew 31% from the year ago level.

從我們第三季的結果可見，各項業務的需求仍然強勁。以公司整體而言，第三季營收較去年創紀錄的第三季成長 16%，非 GAAP 每股盈餘成長 35%。本季我們在工業自動化業務確實遭遇供應鏈瓶頸加劇，導致出貨未能滿足需求。Sanjay 將更詳細說明，但我們預期這些限制將延續至第四季。儘管如此，2021 年前 9 個月公司整體營收較去年同期成長 19%，非 GAAP 每股盈餘成長 31%。

In each of our businesses, we are riding long-term secular trends that we expect will drive revenue and earnings growth for years to come. In our Test businesses, the unit growth and complexity drivers that power these markets continue unabated. For example, our semiconductor test business grew 18% through Q3, while -- with SOC leading the charge growing 22%. Sales continue to be dominated by our UltraFLEX product line which is well aligned to the performance requirements of the growing compute and mobility markets.

在我們各項業務中，我們正乘著長期結構性趨勢的浪潮，我們預期這些趨勢將在未來多年推動營收與獲利成長。在測試業務方面，驅動這些市場的出貨量成長與複雜度提升因素持續不減。例如，我們的半導體測試業務截至第三季成長 18%，其中 SOC 領軍成長 22%。營收仍以我們的 UltraFLEX 產品線為主，其與日益成長的運算與行動市場之效能需求高度契合。

Additionally, sales of our Eagle Test Systems more than doubled in the 9-month period as Automotive and Industrial Test markets have also rapidly expanded. Eagle's unique architecture hits the sweet spot in these markets by balancing high precision with the stress testing needed for these demanding applications.

此外，我們的 Eagle 測試系統在前 9 個月期間的銷售額較去年同期增加逾一倍，因汽車與工業測試市場也快速擴張。Eagle 獨特的架構在這些市場切中要點，能在高精度與這些嚴苛應用所需的壓力測試之間取得平衡。

Within SOC, there's been a clear shift this year to higher demand from the compute, automotive and industrial markets. While mobility is still the largest subsegment of SOC and growing, it has dropped from the high 50% range of the SOC test market in recent years to the high 40% range this year. Over the midterm, we expect mobility will remain the largest SOC submarket and continue to grow but we also expect compute to grow at a faster rate, while automotive should remain at its current elevated levels.

在 SOC 領域，今年明顯轉向來自運算、汽車與工業市場的更高需求。雖然行動仍是 SOC 最大的子區隔且持續成長，但其占 SOC 測試市場的比重已從近年約 50% 後段降至今年約 40% 後段。在中期來看，我們預期行動仍將是最大的 SOC 子市場並持續成長，但也預期運算將以更快速度成長，而汽車則應維持在目前較高的水準。

For the last decade or so, mobility has made rapid annual advances in semiconductor complexity that has enabled the advancement of smartphone sophistication. The refresh pace has been much faster than traditional PCs, graphics, automotive and industrial end markets, leading to smartphone ICs rapidly progressing along the complexity scale. This is true in many areas of smartphone silicon, apps processor compute engines, graphics engines, AI engines, image sensors, power management and more. Our leading position in testing these key technologies has driven our growth.

在過去約十年，行動領域在半導體複雜度方面每年快速進步，推動智慧型手機功能日益精進。其更新節奏遠快於傳統 PC、繪圖、汽車與工業終端市場，使得智慧型手機 IC 在複雜度曲線上迅速攀升。這在智慧型手機晶片的許多領域皆為事實，包括應用處理器運算引擎、圖形引擎、AI 引擎、影像感測器、電源管理等。我們在測試這些關鍵技術方面的領先地位，帶動了我們的成長。

At the same time, up until recently, the traditional compute testing market has been relatively flat with slower refresh rates and slower complexity growth. However, the groundwork laid by mobility designs combined with advancing lithography nodes and design tools has enabled new entrants into the chip design space for compute engines. The complexity of these chips whether for laptops, servers, autonomous driving, AI or graphics is incredible and advancing at an accelerated rate. For example, laptop CPUs are now crossing the 30 billion transistor level, which is a huge leap over previous legacy designs.

同時，直到最近為止，傳統運算測試市場相對持平，更新速度較慢、複雜度成長也較慢。然而，行動設計所奠定的基礎，加上先進微影節點與設計工具的進步，使得運算引擎晶片設計領域出現新的參與者。這些晶片的複雜度——無論用於筆電、伺服器、自動駕駛、AI 或圖形——都令人驚嘆，且正以加速的速度演進。例如，筆電 CPU 現在已跨越 300 億電晶體等級，相較於過去的傳統設計是巨大的躍進。

As we've said in the past, increased transistor counts drives increased test time and increased tester demand. We've seen that this year and there's more to come. We're targeting this expanding collection of new players and new designs, leaning heavily into our UltraFLEX family's hardware performance and time-to-market advantages of our software. We've been adding new design wins every quarter. And while development pipelines can be long and these new designs can be speculative, we're confident we'll see growing production business from these wins in the future. It's also notable that the traditional chip suppliers in these markets aren't standing still, they are doubling down on their advanced designs too, which is collectively driving WFE investments higher as applications expand and competition heats up.

如同我們過去所說，電晶體數量增加會帶來測試時間增加，以及對測試機需求增加。我們今年已看到這一點，未來還會更多。我們正鎖定這個由新玩家與新設計所構成、正在擴大的市場集合，並大幅倚重 UltraFLEX 系列的硬體效能，以及我們軟體在縮短上市時間方面的優勢。我們每一季都在新增設計導入（design win）。雖然開發管線可能很長，且這些新設計可能具有投機性，但我們有信心未來將從這些導入案中看到更多量產業務成長。同樣值得注意的是，這些市場中的傳統晶片供應商並未停下腳步，他們也在加碼其先進設計；隨著應用擴大與競爭升溫，整體正推動晶圓廠設備（WFE）投資提高。

We expect this race to lead to higher test TAMs and given the higher performance and faster design to market cycle times, more share gain opportunities for Teradyne over the midterm.

我們預期這場競賽將帶來更高的測試 TAM（可服務市場總量）；鑑於更高的效能與更快的設計到上市週期時間，Teradyne 在中期也將有更多市占提升的機會。

Our System Test segment last year -- sorry, our System Test segment year-to-date sales grew 11% from 2020 and Storage Test continued its multiyear growth trajectory, expanding sales 12% in the same period. Higher capacity HDDs and more complex SOC devices, which require system-level test, are driving this demand. Both trends are expected to continue into the foreseeable future.

我們的系統測試部門——抱歉，我們的系統測試部門年初至今的銷售額較 2020 年成長 11%，而儲存測試也延續其多年成長軌跡，在同一期間銷售額成長 12%。更高容量的 HDD，以及需要系統層級測試的更複雜 SOC 裝置，正在推動這項需求。我們預期這兩項趨勢在可預見的未來都將持續。

At LitePoint, sales were up 24% through 9 months compared with 2020, driven by WiFi 6E production, WiFi 7 R&D demand as well as ultra-wideband. More connected devices demanding more bandwidth while managing growing congestion drive complexity increases in each new WiFi standard and more tests. UWB, on the other hand, is a whole new wireless standard in application space. It's a new proximity detection wireless technology with the future of many promising security applications. We expect these trends to continue and to provide a long-term tailwind to our Wireless Test business.

在 LitePoint，截至前 9 個月的銷售額較 2020 年成長 24%，主要由 WiFi 6E 量產、WiFi 7 研發需求以及超寬頻（UWB）所帶動。更多連網裝置在管理日益擁塞的同時也需要更高頻寬，推動每一代新 WiFi 標準的複雜度提升與更多測試。另一方面，UWB 則是在應用領域中的全新無線標準。它是一種新的近距離偵測無線技術，未來可望支援許多具前景的安全應用。我們預期這些趨勢將持續，並為我們的無線測試業務帶來長期順風。

Shifting to Industrial Automation. Universal Robots revenue grew 50% through the first 9 months of the year, while MiR grew 40% despite supply chain challenges. Each has a unique story. At UR, it's a combination of increasing sales for existing tasks and the expanding number of UR+ offerings, making it easier for customers to deploy our cobots to do new applications. We highlighted welding in our last call, but other examples include screw-driving and palletizing.

轉向工業自動化。Universal Robots 在今年前 9 個月的營收成長 50%，而 MiR 儘管面臨供應鏈挑戰仍成長 40%。兩者各有獨特的故事。在 UR，這是既有任務銷售增加與 UR+ 產品供應數量擴大相結合的結果，讓客戶更容易部署我們的協作機器人（cobots）來執行新的應用。我們在上次電話會議中強調了焊接，但其他例子還包括螺絲鎖付與棧板堆疊。

The UR+ ecosystem is key to expanding these tasks and now totals over 360 products created by over 300 partners, both riding on and broadening the coattails of our UR platform. This is a key advantage in the combination of our organic investments and our UR+ and OEM partners R&D dollars and creativity that's going into expanding the UR platform and it's unmatched.

UR+ 生態系是擴展這些任務的關鍵，目前總計超過 360 項產品，由 300 多家合作夥伴打造，既依託於我們的 UR 平台，也進一步擴大其影響力。這是我們有機投資與 UR+ 及 OEM 夥伴的研發資金與創意相結合所帶來的關鍵優勢，用於擴展 UR 平台，而且無可匹敵。

At MiR, the story is about new products. The MiR250 which was introduced just as COVID hit last March of last year, is now our largest seller by far. This year, we added the MiR Hook to the 250 family to expand its applications into tugging. We've introduced higher payload products such as the MiR 600 and 1350 to expand our footprint in the fast-growing logistics market.

在 MiR，故事重點在於新產品。MiR250 於去年 3 月 COVID 爆發之際推出，如今已成為我們迄今為止最暢銷的產品。今年，我們為 250 系列新增 MiR Hook，以將其應用擴展至牽引（tugging）。我們也推出更高載重的產品，例如 MiR 600 與 1350，以擴大我們在快速成長的物流市場中的版圖。

Unfortunately, with all this good news come supply chain issues that will limit IA growth in 2021 to be between 30% to 40% year-on-year, but demand is strong. The long-term outlook in IA remains very bright. Looking at the capabilities of UR cobots today, we estimate the penetration rate is less than 2% of the servable market. UR's approximate 45% market share puts us clearly in the lead, and we continue to drive R&D and distribution investments to extend our competitive advantages, expand the servable market and drive penetration higher.

不幸的是，伴隨這些好消息而來的是供應鏈問題，將使 2021 年 IA 成長受限，年增率介於 30% 至 40%，但需求依然強勁。IA 的長期展望仍非常光明。以 UR 協作機器人目前的能力來看，我們估計其滲透率不到可服務市場的 2%。UR 約 45% 的市占率使我們明顯領先，我們也持續推動研發與通路投資，以延伸競爭優勢、擴大可服務市場並提高滲透率。

It's a similar story at MiR, where we estimate the autonomous mobile robot penetration is under 3%. The AMR market doesn't have a single dominant player like UR cobots, and we estimate we're close to #2 in the broadly defined market. And like at UR, we're making investments in both the distribution and product level to both reinforce our advantages and extend our product reach.

MiR 也有類似的情況，我們估計自主移動機器人（AMR）的滲透率低於 3%。AMR 市場不像 UR 協作機器人那樣有單一主導者，我們估計在廣義市場中接近第 2 名。並且如同 UR，我們在通路與產品層面都持續投資，以鞏固優勢並延伸產品觸及範圍。

In both IA businesses, the fact that our penetration of today's servable market is low single digits and that the servable market continues to expand each year with product enhancements, sets up a fantastic future. So even with very high growth rates in our IA business, we expect the penetration rates to remain low for many years, sustaining our long-term annual growth forecast of 20% to 35%. In January, we will update you on the outlook for 2022 and our midterm earnings model. Between now and then, we'll be looking at the rate and timing of new semiconductor fab capacity coming online, especially at the more advanced lithography nodes. And we'll also be looking at the rate of adoption of DDR5 as key swing factors.

在兩項 IA 業務中，我們在當前可服務市場的滲透率僅為個位數低段，而可服務市場又會隨著產品強化每年持續擴大，這為未來奠定了極佳的基礎。因此，即使 IA 業務成長率很高，我們預期滲透率在多年內仍將維持偏低，支撐我們長期每年 20% 至 35% 的成長預測。我們將在 1 月更新 2022 年展望以及中期獲利模型。在此之前，我們將關注新增半導體晶圓廠產能上線的速度與時點，特別是在更先進的微影節點。我們也將觀察 DDR5 的採用速度，作為關鍵的變動因素。

In IA, we will be looking at the manufacturing output expansion, onshoring trends and PMIs in our principal geographies as tailwinds for continued robust growth. On the other hand, in both markets, supply chain bottlenecks could slow certain industries and become a headwind to growth demand. Short-term demand is influenced by many factors, but we manage our business aligned to the long-term trends. The trend of growing prevalence of increasingly complex semiconductors and a myriad of applications drives our semiconductor business and investments. The trend of new increasingly smart, cost-effective automation in a world with labor scarcity and onshoring challenges drives our IA business and investment strategy. These systemic long-term trends paint an exciting future for Teradyne.

在 IA 方面，我們將把製造產出擴張、回流（onshoring）趨勢，以及主要地區的採購經理人指數（PMI）視為持續強勁成長的順風因素。另一方面，在兩個市場中，供應鏈瓶頸可能拖慢某些產業，並成為需求成長的逆風。短期需求受多種因素影響，但我們的業務管理與長期趨勢保持一致。日益普及的更複雜半導體與各式各樣的應用趨勢，推動我們的半導體業務與投資。在勞動力短缺與回流挑戰的世界中，愈來愈智慧且具成本效益的自動化趨勢，推動我們的 IA 業務與投資策略。這些系統性的長期趨勢為 Teradyne 描繪出令人振奮的未來。

With that, I'll turn it over to Sanjay.

接下來我把時間交給 Sanjay。

Thanks, Mark, and hello, everyone.

謝謝你，Mark，各位大家好。

In my remarks, I'll review our Q3 financial results, discuss our supply line strategy in this challenging environment, provide Q4 guidance and comment on our full year financial outlook at the midpoint of our Q4 guidance.

在我的發言中，我將回顧我們第三季的財務結果，討論在這個充滿挑戰的環境下我們的供應線策略，提供第四季指引，並就以第四季指引中點推算的全年財務展望發表評論。

To the financial headlines for Q3. Our third quarter sales were $951 million was near the high end of guidance driven by strength in Semi Test and Wireless Test. Gross margin in the quarter was approximately 60%. Our non-GAAP operating expenses were $242 million or 25.5% of revenue. The favorability in OpEx drove a non-GAAP operating margin of approximately 35% and non-GAAP EPS of $1.59. A few more components of third quarter data. Our tax rate, excluding discrete items, was 14.8% on both a GAAP and non-GAAP basis. Non-GAAP diluted share count was approximately 176 million. We had two 10% customers.

先看第三季的財務重點。我們第三季銷售額為 9.51 億美元，接近指引區間上緣，主要由半導體測試與無線測試的強勁表現帶動。本季毛利率約為 60%。我們的非 GAAP 營業費用為 2.42 億美元，約占營收 25.5%。營業費用的有利表現帶動非 GAAP 營業利益率約 35%，非 GAAP 每股盈餘（EPS）為 1.59 美元。再補充幾項第三季數據。我們的稅率（不含一次性項目）在 GAAP 與非 GAAP 基礎上皆為 14.8%。非 GAAP 稀釋後股數約為 1.76 億股。我們有兩家客戶的營收占比達 10%。

Looking at the results from a business unit perspective. Semi Test revenue of $688 million was up 16% from Q3 '20. SOC revenue was $575 million, up 28%, driven by strength in applications processors, RF, industrial and automotive applications. Automotive and industrials doubled revenue year-over-year. Memory revenue was the second highest in history at $113 million, but down 21% from Q3 of last year's record.

從事業部角度來看結果。半導體測試營收為 6.88 億美元，較 2020 年第三季成長 16%。SOC 營收為 5.75 億美元，成長 28%，主要受惠於應用處理器、RF、工業與車用應用的強勁需求。車用與工業營收年增翻倍。記憶體營收為 1.13 億美元，為歷史第二高，但較去年創紀錄的第三季下滑 21%。

FLASH final test demand was the strongest segment on handset and SSD end market demand. System Test group had revenue of $103 million, which was down 13% year-over-year. Recall Storage is the largest business in this segment and has lumpy shipments. While Storage Test will still grow more than 15% for the year, sales, including HDD and SLT declined to $56 million on the timing of shipments in Q3.

FLASH 終測需求為最強勁的區塊，受手機與 SSD 終端市場需求帶動。系統測試事業群營收為 1.03 億美元，年減 13%。其中 Storage 是該區塊最大的業務，出貨具有波動性。雖然 Storage Test 全年仍將成長超過 15%，但由於第三季出貨時點影響，包含 HDD 與 SLT 在內的銷售下滑至 5,600 萬美元。

Defense and aerospace and production board test combined grew 10% year-on-year to $47 million. At LitePoint, revenue of $69 million was up 70% from prior year due to early success of our new WiFi 7 product, continued strength in 4G cellular and UWB.

國防與航太以及生產板測試合計年增 10% 至 4,700 萬美元。在 LitePoint，營收為 6,900 萬美元，較去年成長 70%，主要因我們新 WiFi 7 產品的早期成功，以及 4G 行動通訊與 UWB 的持續強勁表現。

Now to Industrial Automation. As in July, given COVID shutdowns had impaired the UR business in 2020, I'll provide revenue metrics comparing Q3 '21 results with both Q3 '20 and Q3 '19. Industrial Automation revenue of $91 million was up 32% from both Q3 '19 and Q3 '20. North America delivered the highest revenue growth from last year, but all regions expanded year-on-year. As Mark noted, supply issues, primarily semiconductors, reduced our IA shipments in the quarter. UR sales were $78 million in Q3, up 46% year-over-year and 31% over Q3 '19. MiR sales were $13 million, up 27% from Q3 '20 and 35% from Q3 '19. IA was about breakeven in the quarter and full -- breakeven in the quarter and full year, we expect low single-digit profitability. As we've noted before, we continue our strategy of investing during this high-growth era while maintaining gross margins to enable mid-20s operating profit in the future.

接著談工業自動化。如同 7 月所述，鑑於 2020 年 COVID 停工影響了 UR 業務，我將提供營收指標，將 2021 年第三季結果同時與 2020 年第三季及 2019 年第三季比較。工業自動化營收為 9,100 萬美元，較 2019 年第三季與 2020 年第三季皆成長 32%。北美較去年呈現最高的營收成長，但所有地區皆年增。如 Mark 所提，供應問題（主要是半導體）降低了我們本季 IA 的出貨量。UR 第三季銷售額為 7,800 萬美元，年增 46%，較 2019 年第三季成長 31%。MiR 銷售額為 1,300 萬美元，較 2020 年第三季成長 27%，較 2019 年第三季成長 35%。IA 本季約為損益兩平，而全年我們預期為低個位數獲利。如我們先前所述，我們將持續在這個高成長時期進行投資，同時維持毛利率，以在未來實現 20% 多的營業利益率。

Shifting to supply. We continue to deal with numerous supply constraints across the company. While semiconductor shortages are well reported, we're also seeing delays in mechanical parts and logistics, all exacerbated by rolling COVID-related shutdowns or labor shortages. We expect these issues will continue through the first half of 2022. Despite these issues, we've been able to deliver record shipments and a big part of that performance is a result of the supply line management, operations teams and engineering teams working with our supply chain and contract manufacturing partners.

轉到供應面。我們仍在全公司範圍內面臨眾多供應限制。雖然半導體短缺已廣為報導，但我們也看到機械零件與物流的延遲，而這些情況都因持續滾動式的 COVID 相關停工或勞動力短缺而加劇。我們預期這些問題將持續到 2022 年上半年。儘管如此，我們仍能交付創紀錄的出貨量，而這項表現很大一部分歸功於供應線管理、營運團隊與工程團隊，與我們的供應鏈及委外製造合作夥伴共同努力的成果。

We view our operational business model and execution against it as a core competence. Our gross margin performance over the last 10 years displays the financial value of this model.

我們將自身的營運商業模式以及對該模式的執行視為核心能力。過去 10 年的毛利率表現展現了此模式的財務價值。

In our test portfolio, our execution has kept most of our tester lead times within the range that meet customers' needs to expand their production capacity in this dynamic environment. The significance of this lead time performance is that customer orders more closely reflect true test demand. So while a bit counterintuitive, we feel that maintaining short lead times are a more accurate indicator of test demand with lower risk than holding orders with lead times far beyond chip manufacturing cycle times. Of course, our supply line and operations model isn't static. We began adding resiliency through both geographic and supplier diversity prior to COVID. Those efforts have accelerated over the last 20 months.

在我們的測試產品組合中，我們的執行使大多數測試機的交期維持在能滿足客戶在此動態環境中擴充產能需求的範圍內。此交期表現的重要性在於，客戶訂單能更貼近真實的測試需求。因此，雖然有點違反直覺，我們認為維持較短交期，比起持有交期遠超過晶片製造週期時間的訂單，更能作為測試需求的精準指標且風險更低。當然，我們的供應線與營運模式並非一成不變。在 COVID 之前，我們就已開始透過地理分散與供應商多元化來提升韌性。過去 20 個月，這些努力已加速推進。

This work is paying dividends in the current environment, and we'll continue to invest to harden our supply chain further. Our lead time performance is an example of our resilience and execution. Another example is the ability to scale to significantly increase demand. Auto and industrial sales have more than doubled year-over-year. While we're not perfectly aligned to all customer requested delivery dates, we are managing through delivery issues in a reasonable manner. The value of these efforts can also be seen in the operating leverage in our gross margin line.

這些工作在目前環境中正帶來成效，我們也將持續投資以進一步強化供應鏈。我們的交期表現就是韌性與執行力的例證。另一個例子是能夠擴充規模以因應顯著增加的需求。汽車與工業銷售年增已超過一倍。雖然我們未能與所有客戶要求的交貨日期完全一致，但我們正以合理方式管理並克服交付問題。這些努力的價值也可從我們毛利率項目的營運槓桿中看出。

In IA, we've seen lead times extend from our normal 1 to 2 weeks to 4 to 6 weeks for some products. While this is a challenge with the business growing so quickly and ongoing industry supply issues, we've already seen the positive impact of our work in material sourcing and manufacturing cycle times. We expect to bring lead times back to model over the next 2 to 3 quarters.

在 IA 方面，我們看到部分產品的交期從正常的 1 到 2 週延長至 4 到 6 週。在業務快速成長與產業供應問題持續的情況下，這確實是一項挑戰，但我們已看到在材料採購與製造週期時間方面工作的正面影響。我們預期在未來 2 到 3 個季度內，將交期拉回到模型水準。

Shifting to the balance sheet and cash flow. Our cash and marketable securities at the end of the quarter totaled $1.45 billion. We had $493 million in free cash flow in the quarter, and through 9 months, we've spent $103 million on CapEx, and we expect we'll spend $148 million for the full year. We spent $210 million and $16 million on buybacks and dividends, respectively. Year-to-date, we've repurchased 3.3 million shares for $406 million at an average price of $123.53. In 2021, we expect to return over 80% of our free cash flow to shareholders. And from 2015, when we began repurchasing shares, we've returned 85% of our free cash flow to owners.

接著談資產負債表與現金流。本季末我們的現金與有價證券合計為 14.5 億美元。本季自由現金流為 4.93 億美元；累計前 9 個月，我們在資本支出（CapEx）上已支出 1.03 億美元，並預期全年將支出 1.48 億美元。我們分別在庫藏股回購與股利上支出 2.10 億美元與 1,600 萬美元。年初至今，我們以平均價格 123.53 美元回購 330 萬股，總額 4.06 億美元。在 2021 年，我們預期將把超過 80% 的自由現金流回饋給股東。自 2015 年開始回購股票以來，我們已將 85% 的自由現金流回饋給所有人（股東）。

Regarding our convertible debt, $302 million of principal was paid in the first 9 months to the convertible bondholders ahead of maturity. By mid-December, bondholders will have converted approximately $343 million, leaving a face value of $117 million.

關於我們的可轉換公司債，前 9 個月我們已在到期前向可轉債持有人償付 3.02 億美元本金。到 12 月中旬，債券持有人將轉換約 3.43 億美元，剩餘面額為 1.17 億美元。

Now to our outlook for Q4. Sales in Q4 are expected to be between $820 million and $900 million, with non-GAAP EPS in a range of $1.14 to $1.40 on 174 million diluted shares. Fourth quarter guidance excludes the amortization of acquired intangibles and noncash imputed interest on the convertible debt. Our guidance assumes no significant changes, positive or negative, in the availability of materials and assumes that we won't see additional pandemic-related issues. Fourth quarter gross margins are estimated at 59% to 60%. OpEx is expected to run at 28% to 31% of fourth quarter sales. The non-GAAP operating profit at the midpoint of our third quarter guidance is 30%.

接著是我們對第四季的展望。第四季銷售額預期介於 8.20 億至 9.00 億美元之間，以 1.74 億股稀釋後股數計算，非 GAAP 每股盈餘（EPS）預期介於 1.14 至 1.40 美元。第四季指引不包含併購取得無形資產的攤銷，以及可轉債的非現金推算利息。我們的指引假設材料供應可得性不會出現重大變化（無論正面或負面），並假設不會再出現額外的疫情相關問題。第四季毛利率估計為 59% 至 60%。營業費用（OpEx）預期約為第四季銷售額的 28% 至 31%。以我們第三季指引的中點計算，非 GAAP 營業利益率為 30%。

Regarding OpEx for the full year. We spent a bit lower than planned in Q3, and we expect the full year OpEx will be about $980 million, up 17% from 2020. At the midpoint of our guidance, 2021 will be another year of growth in both revenue and EPS, while sales growing -- or sorry, with sales growing 18% to $3.7 billion and non-GAAP EPS growing to $5.88, up 27%. Gross margin for the full year should be approximately 59.5%, up from 57.2% in 2020, reflecting the ramp of new products, product mix and operating leverage, offsetting component and logistics cost increases.

關於全年營業費用（OpEx）。我們第三季支出略低於原先規劃，預期全年 OpEx 約為 9.80 億美元，較 2020 年增加 17%。以指引中點來看，2021 年將是營收與 EPS 再度成長的一年；同時，銷售額成長 18% 至 37 億美元，非 GAAP EPS 成長至 5.88 美元，年增 27%。全年毛利率應約為 59.5%，高於 2020 年的 57.2%，反映新產品放量、產品組合與營運槓桿，並抵銷零組件與物流成本上升的影響。

Our 2021 non-GAAP operating profit rate is expected to be about 33%, up from last year's 30%. Our full year tax rate is expected to be 14.8%. These results put us comfortably in the range of our 2024 earnings model this year. We'll update the model on our regular cadence in January.

我們 2021 年非 GAAP 營業利益率預期約為 33%，高於去年的 30%。全年稅率預期為 14.8%。這些結果讓我們今年能穩健落在 2024 年獲利模型的區間內。我們將在 1 月依既定節奏更新該模型。

The breadth of our customer buying in 2021 is also broader than last year. In 2020, we had 1 customer that drove 25% of sales. In 2021, we do not expect to have any customer larger than 20% of our yearly revenue. This reflects the trend Mark noted about 2021's higher growth in compute, auto and industrial demand compared with mobility demand in Semi Test. IA growth of over 30% year-over-year continues to diversify our revenue. This growth is expected to continue over the midterm and become a larger portion of our revenue.

我們 2021 年客戶採購的廣度也比去年更廣。在 2020 年，我們有 1 位客戶貢獻了 25% 的銷售額。在 2021 年，我們預期不會有任何單一客戶占全年營收超過 20%。這反映了 Mark 所提到的趨勢：相較於半導體測試（Semi Test）中的行動裝置需求，2021 年在運算、汽車與工業需求的成長更高。IA 年增超過 30% 的成長持續使我們的營收更加多元化。預期這項成長在中期仍將延續，並成為我們營收中更大的一部分。

In summary, we expect to end the year with another quarter of strong year-on-year revenue and EPS growth. On a full year basis, we'll exceed our midterm targets on stronger-than-expected demand on our test businesses, continued high growth in IA and excellent execution across the company. While we don't have a clear picture of 2022 yet, we're confident that the long-term industry trends powering our test and IA businesses remain firmly in place.

總結來說，我們預期今年將以另一個營收與 EPS 年增強勁的季度作結。以全年來看，受惠於測試業務需求強於預期、IA 持續高速成長，以及全公司卓越的執行力，我們將超越中期目標。雖然我們對 2022 年尚未有清晰全貌，但我們有信心，推動我們測試與 IA 業務的長期產業趨勢仍然穩固存在。

With that, I'll turn the call back to Andy.

接下來，我把電話會議交回給 Andy。

Thanks, Sanjay. Patrice, we would now like to take some questions. (Operator Instructions)

謝謝你，Sanjay。Patrice，我們現在想開始回答一些問題。(接線員指示)

(Operator Instructions) Your first question comes from Atif Malik with Citi.

(接線員指示) 你們的第一個問題來自 Citi 的 Atif Malik。

Good job in the tough supply environment. And Mark, if I look at some of the recent ARM-based notebook processors, the transistor count is growing 2 to 4x versus prior generation. You commented mobility to grow and compute, auto, analog to grow faster than mobility in the midterm, understand you guys generally talk about next year outlook in January. How confident do you feel about your SOC growth next year?

在艱難的供應環境下做得很好。Mark，如果我看一些近期採用 ARM 架構的筆電處理器，電晶體數量相較前一代成長了 2 到 4 倍。你提到行動裝置會成長，而在中期運算、汽車、類比的成長會快於行動裝置；我理解你們通常會在 1 月談明年的展望。你對於明年 SoC 成長的信心有多高？

Well, I think at this point, pretty confident. I mean we look at the trends that I just -- you cited and I cited and see that there's very little standing in the way of sort of this increased growth, but the caveat I would mention is there's a lot of supply chain bottlenecks in the system. So if end products get bottlenecked, that can slow down the unit volume demand for new semiconductors. And the other thing we're looking at is when do these new nodes really come online in terms of capacity. All that $90 billion of WFE that was put in place this year hasn't yet had a single impact on test. That's all to come. But it kind of depends on when do those additional 5- and 3-nanometer fabs come online. That is kind of a big swing factor in the calendar year 2022 as to how our growth will chunk up. And that's kind of why we wait until January because we don't get a good enough visibility on that right now.

嗯，我想在這個時間點，我們相當有信心。我的意思是，我們看我剛才——你提到的、我也提到的那些趨勢，會看到幾乎沒有什麼能阻擋這種成長加速；但我想提醒的一點是，系統裡存在許多供應鏈瓶頸。所以如果終端產品被卡住，可能會放緩對新半導體的單位出貨量需求。另外一件我們在觀察的是，這些新節點的產能究竟何時真正上線。今年投入的那 900 億美元 WFE，對測試端還沒有產生任何影響。那都還在後面。但這在某種程度上取決於那些額外的 5 奈米與 3 奈米晶圓廠何時投產。這是 2022 年曆年中，決定我們成長如何分布的一個很大的擺動因素。也正因如此我們會等到一月，因為目前我們對此的能見度還不夠好。

Great. And Sanjay, as a follow-up, can you talk about concentration within mobility and compute customer and impact on long-term gross margins? There have been talks about price discounts with equipment suppliers by Tier 1 foundry.

很好。Sanjay，作為追問，你能談談在行動裝置與運算客戶中的集中度，以及對長期毛利率的影響嗎？市場上一直在談 Tier 1 晶圓代工廠向設備供應商要求價格折扣。

Yes. As I stated in my prepared remarks, we won't have any customer above 20%. So it's a broader breadth of customers. And regarding gross margins, in throughout the year, we've improved gross margins. And as I said in earlier calls, really driven by a couple of key test systems coming online and we're shipping in volume that have come down the cost curve. We've seen some benefits to product mix shift as well as our operating leverage offsetting the component costs and logistic cost increases. From a look forward and sustainability of that gross margin, obviously, we'll give an update to our earnings model in January. But I see the second half of our gross margin performance kind of going into the first half of 2022.

是的。如我在事先準備的發言中所說，我們不會有任何單一客戶占比超過 20%。因此客戶基礎會更為多元、覆蓋更廣。至於毛利率，我們在全年期間持續改善毛利率。而且如我在先前的電話會議中提到，主要是由幾套關鍵測試系統上線所帶動；我們以量產出貨，並且成本曲線已經下降。我們也看到一些產品組合轉移的效益，以及我們的營運槓桿抵消了零組件成本與物流成本的上升。展望未來以及毛利率的可持續性，當然我們會在一月更新我們的獲利模型。但我認為我們下半年毛利率的表現，會延續到 2022 年上半年。

Your next question comes from Mehdi Hosseini with SIG.

下一個問題來自 SIG 的 Mehdi Hosseini。

Yes, the first one has to do with your largest customer. And given your commentary in terms of revenue mix, it seems to me that that particular customer is going to be down like 5% to 6%. In that context, should we assume a return to growth in '22, and this is a trend that has happened over the past several years? And should that happen again -- Should that pattern happen again into '22? And I have a follow-up.

是的，第一個問題與你們最大客戶有關。依照你們對營收組合的評論，在我看來，那位特定客戶可能會下滑大約 5% 到 6%。在這個背景下，我們是否應該假設 2022 年會回到成長，而這也是過去幾年一直發生的趨勢？而這種模式是否會在 2022 年再次出現——也就是說，這個模式是否會在 2022 年延續？我還有一個追問。

Yes. We, of course, maybe can't talk about any individual customer and what they might do or not do in the future. So that I have to leave aside. But I just point out in your first point that, yes, our largest customer is dropping below 20% in a growing revenue year. So the amount that you might think compute that they're falling has to be taken against our numerator of higher revenue.

是的。我們當然可能無法談論任何單一客戶，以及他們未來可能會做或不會做什麼。所以那部分我必須先擱置。但我想指出你第一點：是的，在營收成長的一年裡，我們最大客戶的占比正在降到 20% 以下。因此，你可能認為他們在運算端下滑的幅度，必須放在我們分子——也就是更高的營收——的背景下來看。

Got it. And then I wanted to follow up to the question that came up 3 months ago when we were looking into your market share in the compute. Can you update us where you are with that market share in 2021? And as hyperscalers ramp their own ARM-based CPUs, how will your market share change over the next 1 or 2 years?

了解。接著我想追問三個月前出現的那個問題，當時我們在看你們在運算領域的市占率。你能更新一下你們 2021 年的市占率狀況嗎？以及當超大規模業者（hyperscalers）加速導入他們自家的 ARM 架構 CPU 時，你們的市占率在未來 1 到 2 年會如何變化？

Well, like the total SOC market, year-to-year market share is very volatile. It depends on whose customers are buying what in any given year. So what our market share might be in any given year, it can -- in a submarket like compute, it can swing 20 points year-to-year depending on who's buying. So for this year, it tends to be a very good compute year for us. Our compute market share is up in, I would say, close to what our average share is in SOC this year. But I would say that it's not steady Eddie. It's going to be pretty volatile year-to-year, it has been in the past. It probably will be going forward, too.

嗯，就像整體 SoC 市場一樣，逐年的市占率波動非常大。這取決於在任何特定年度，哪些客戶在買什麼。所以我們在任何一年可能呈現的市占率——在像運算這樣的子市場——會因為誰在採購而出現逐年 20 個百分點的擺動。因此就今年而言，對我們來說是非常好的一個運算年。我會說，我們的運算市占率上升，接近我們今年在 SoC 的平均市占水準。但我也會說，這不是那種穩穩當當、年年固定的狀態。它會逐年相當波動，過去一直如此。未來大概也會是這樣。

I know we're supposed to ask only 2 questions, but just a quick follow-up. I think what I'm trying to understand is I think on the GPU side, it's pretty clear that your competitor has dominated. And assuming that that work to remain unchanged, I think the incremental change to the compute is all driven by ARM-based. And I was just trying to better understand how you look at your competitive position as these new chips coming to the market addressing the compute end market?

我知道我們應該只問兩個問題，但我再快速追問一下。我想我試圖理解的是，我認為在 GPU 端，很明顯你們的競爭對手已經占據主導地位。而假設那部分的格局維持不變，我認為運算端的增量變化就完全是由 ARM 架構所驅動。我只是想更清楚了解，當這些新晶片進入市場、瞄準運算終端市場時，你們如何看待自己的競爭地位？

We -- look, we're very pleased and confident with our progress in ARM-based compute and design-ins in that realm, but I would say that it's not -- there's the tester market for compute also driven by more traditional x86 demand as well. You've got a couple of suppliers there that are not standing still, as I mentioned, and are upping their kind of complexity growth curve. And so there will be certainly a lot of growth there as well, I believe.

我們——你看，我們對於在 ARM 架構運算領域的進展與設計導入（design-ins）感到非常滿意且有信心；但我會說，這並不是——運算的測試機市場也同樣受到較傳統的 x86 需求所驅動。那裡有幾家供應商並沒有停下腳步，如我提到的，他們正在提高其複雜度的成長曲線。因此我相信，那裡也一定會有相當多的成長。

Your next question comes from Toshiya Hari, Goldman Sachs.

下一個問題來自高盛（Goldman Sachs）的 Toshiya Hari。

Congrats on the strong execution. I had 2 questions as well. My first one is on the supply constraints, maybe for Mark, maybe for Sanjay. Just curious how significant the headwinds were for IA in Q3 and what's embedded in your Q4 guidance? If you can share that, that would be helpful. And I just wanted to confirm that there was little to no impact on your Semi Test business. And then on gross margins, similarly, you came in at the high end of your guided range, but was there any impact on your profitability in the quarter from supply chain shortages?

恭喜你們強勁的執行表現。我也有兩個問題。第一個是關於供應受限，可能問 Mark，也可能問 Sanjay。我想了解 IA 在第三季面臨的逆風有多大，以及你們第四季指引中包含了哪些假設？如果你們能分享，會很有幫助。另外我也想確認一下，你們的半導體測試（Semi Test）業務幾乎沒有受到影響，或完全沒有受到影響。再來關於毛利率，同樣地，你們落在指引區間的高端，但本季供應鏈短缺是否對你們的獲利能力造成任何影響？

Sure. So I'll take those. So from an Industrial Automation perspective, year-to-date, we've grown 40% in IA, and Q4 demand is high. If we can't supply at all, that's why we've -- I think it's going to be growth year-on-year of about 30% to 40%. And so predominantly, it's in semiconductors. And I expect we'll be out of the supply chain crunch given our visibility in Q2 or Q3 for IA. So that's the IA side. And then from a supply chain perspective on the test side, we've seen the supply chain tightening on the test portfolio quarter-over-quarter, where we don't think it's going to be abated until the end of Q2 of 2022. And again, mainly semiconductor parts. And really, we see that coming back online in the second half, really tied to the wafer and the substrate capacity coming online.

好的。那我來回答。從工業自動化（Industrial Automation）的角度來看，今年截至目前 IA 成長了 40%，而第四季需求很高。如果我們完全無法供貨，這也是為什麼我們——我認為年對年成長大約會是 30% 到 40%。而且主要是在半導體相關。就我們目前的能見度來看，我預期 IA 的供應鏈緊縮狀況會在第二季或第三季解除。以上是 IA 的部分。接著從測試端的供應鏈角度，我們看到測試產品組合的供應鏈逐季收緊，我們認為要到 2022 年第二季末才會緩解。同樣地，主要是半導體零組件。而且我們確實看到它會在下半年恢復供應，主要與晶圓與基板產能上線密切相關。

And then from a gross margin perspective, we've been managing through the component increases and logistics increases in cost. And we've -- as I've said earlier, we've had favorable product mix. And as our volumes or our revenues are higher, we're gaining operating leverage along with coming down the cost curve of our new products that we've introduced late last year. So we're managing through it, and it's true well-publicized component cost increases.

接著從毛利率的角度來看，我們一直在因應元件與物流成本上升的情況。而且我們——如我先前所說——有有利的產品組合。隨著我們的出貨量或營收提高，我們正在獲得營運槓桿，同時也隨著我們在去年底推出的新產品沿著成本曲線下降。所以我們正在加以管理，而這確實是外界廣泛報導的元件成本上升。

Got it. That's super helpful. And then as my follow-up, Mark, I wanted to ask about your Eagle Test business. In your prepared remarks, you noted that the business is up more than 2x year-to-date. I guess historically, like many other parts of your business, I think the Eagle Test has been quite cyclical. You would be up for 1 year, 1.5 years, 4 to 6 quarters and then down a little bit as customers digest their test capacity. Based on what you said, it seemed like you're expecting 2022 to be another strong year. I guess the question is what's different this time as you think about Eagle Test into '22. I realize there's complexity growth, but you could argue there's been complexity growth for a very long time. So just curious how different this cycle could be relative to past cycles.

了解。這非常有幫助。接著作為追問，Mark，我想問一下你們的 Eagle Test 業務。在你準備好的發言中，你提到該業務年初至今成長超過 2 倍。我想從歷史上看，和你們業務的其他部分一樣，Eagle Test 一直相當具週期性。通常會成長 1 年、1.5 年、約 4 到 6 個季度，然後在客戶消化其測試產能後又回落一些。根據你所說的內容，似乎你預期 2022 年會是另一個強勁的一年。所以問題是，當你看 Eagle Test 進入 2022 年時，這次有什麼不同？我知道複雜度在成長，但你也可以說複雜度成長已經持續很久了。所以只是好奇，這一輪週期相較於過去的週期，可能會有多不一樣。

Right. Good question. And in addition, automobile unit volume isn't near its historical peak either. So how could this thing keep going beyond the normal 6-quarter surge in automotive, which is you're right, again, that's kind of the traditional pattern. But I think what we see happening and makes us believe this will persist at least through 2022 is there's a lot of silicon refresh going on that's new in the automotive space. New -- kind of racing to get new, more current generation lithography node silicon into automotive design because the legacy lines out there are hard to get at. Chip suppliers are trying to obsolete those fabs. And so the automotive customers don't have as much, I would say, power in this frothy demand environment on the semiconductor supply side. And so the semiconductor suppliers are kind of same, get with the program, move to more advanced nodes. And a little bit of that's happening, which is what -- that means complexity, yield issues, and that means a little more test than you might expect. And I think that's what's giving us a different view this time.

對。好問題。另外，汽車的單位出貨量也還沒回到歷史高峰。那麼這件事怎麼可能超越汽車領域通常 6 個季度的衝高期——你說得沒錯，那確實是傳統模式。但我認為我們看到、也讓我們相信至少會延續到 2022 年的原因是：汽車領域正在發生大量新的矽晶更新換代。大家正在競相把更新、更先進製程節點的矽晶導入汽車設計，因為現有的舊產線很難取得產能。晶片供應商正試圖讓那些晶圓廠退役。因此在半導體供應端需求火熱的環境下，汽車客戶我會說沒有那麼大的議價能力。所以半導體供應商基本上是同樣的態度：跟上進度，轉向更先進的節點。這種情況正在發生一些，而這代表複雜度、良率問題，也意味著測試需求會比你原本預期的更多一些。我想這就是這次我們看法不同的原因。

Your next question comes from John Pitzer, Credit Suisse.

下一個問題來自瑞士信貸（Credit Suisse）的 John Pitzer。

Congratulations on the solid results. Mark, I want to go back to increasing test times in the Semi Test business. You've done a good job kind of helping us understand complexity and transistor count. I'm kind of curious, as we move from a world of sort of general purpose compute to one of more optimized silicon, you're going to move from a world where you're testing huge volumes in one device to smaller volumes across multiple devices. What does that do for test efficiency at your customers and hence, test times as that trend takes hold?

恭喜你們交出穩健的成果。Mark，我想回到半導體測試（Semi Test）業務中測試時間增加的議題。你已經很清楚地幫助我們理解複雜度與電晶體數量的影響。我有點好奇，當我們從通用運算的世界走向更最佳化的矽晶時，你會從在單一裝置上測試巨大出貨量，轉變為在多個裝置上測試較小出貨量。當這個趨勢成形時，這會如何影響你們客戶的測試效率，進而影響測試時間？

It is a -- building, let's say, 100 billion transistors on 1 chip versus 4, to give an example, isn't equivalent test time. The 4-chip version is likely -- these are going to be rough rules of thumb, but let's say, 25% more test intensive than the single chip design because there's a premium on sort of known good die testing to -- when you put those together in an advanced package. And then the advanced package itself has more potential defect failure modes that need to be tested. So that's one thing around the whole chiplet, multi-chip package thing.

也就是說——舉例來說——把 1,000 億個電晶體做在 1 顆晶片上，和做在 4 顆晶片上，測試時間並不等同。4 晶片版本很可能——這些只是粗略的經驗法則——會比單晶片設計多約 25% 的測試強度，因為在把它們整合到先進封裝時，對「已知良品裸晶（known good die）」測試有額外要求。而且先進封裝本身也有更多潛在的缺陷失效模式需要測試。所以這是關於整個 chiplet、多晶片封裝趨勢的一個面向。

The other thing, though, that's happening, and this is really going to become prevalent at 3-nanometer and beyond is the move from FinFET to gate-all-around transistor architectures. And if you remember, when the world moved from planar to FinFET, it was back in 2012, '13, '14 era, that drove -- if you go back and look at the history of sort of the test market, that drove incremental test intensity and complexity in the market. And we're headed for another one of those with gate-all-around. There's going to be new defect modes, new kind of test intensity boosts coming from this new architecture on the transistor. So even at equivalent transistor counts, we're going to see a little bit more test intensity because of that. So these 2 trends of chiplets and gate-all-around and 3-nanometer, that's probably all the 2023 and beyond story given where 3-nanometer is right now, but it's coming.

不過另一件正在發生的事——而且在 3 奈米及更先進節點將會變得普遍——是從 FinFET 轉向環繞閘極（gate-all-around, GAA）電晶體架構。如果你還記得，當產業從平面（planar）轉向 FinFET，大約是在 2012、2013、2014 年那段期間，這帶動了——如果你回頭看測試市場的歷史——市場測試強度與複雜度的增量提升。而我們正要迎來另一波由 GAA 帶動的類似變化。這種新的電晶體架構會帶來新的缺陷模式，以及新的測試強度提升。因此即使在相同的電晶體數量下，我們也會因為這個因素看到測試強度略為增加。所以 chiplet、GAA 與 3 奈米這兩個趨勢，考量到目前 3 奈米的進度，可能主要是 2023 年及以後的故事，但它正在到來。

That's helpful. And that's my second question, Mark, just going back to your largest customer, I'm kind of curious if you can help me better understand the diversity of business with that customer. I mean clearly, it's been mobility-led for the last several years. Now you've got them doing more in the compute space. You've talked about in the past the complexity around these AirTags. Are you seeing a meaningful diversification of demand drivers at that customer? And can you help me -- because I just don't know how fungible is your -- is their test capacity across those different product families?

這很有幫助。接著是我的第二個問題，Mark，回到你們最大的客戶，我有點好奇你能否幫我更好地理解你們與該客戶之間業務的多樣性。很明顯，過去幾年主要是由行動裝置帶動。現在你們也看到他們在運算領域做得更多。你過去也談過 AirTags 相關的複雜度。你是否看到該客戶的需求驅動因素出現有意義的多元化？以及你能否幫我——因為我不太清楚——你們的（或他們的）測試產能在不同產品族群之間的可互換性到底有多高？

Yes, I'm not going to be able to talk too much about the sort of breadth there. I think you can imagine it, and you've rattled off some examples. In terms of fungibility, they're pretty fungible to testers across what's used for compute engine in a phone versus a compute engine in a laptop or a desktop or anything else. The one that's a little bit unique is when you get into things like, the RF type products, of course, have a bit of a different architecture, power management type products have a bit of a different architecture. And so those testers tend to be a little more unique. But we've talked about in the past that -- I think people have looked at us and said, well, in a world where cell phones are kind of plateauing, doesn't that -- unit growth of cell phones doesn't that that portends some kind of slowing for us.

是的，我沒辦法就那個廣度談太多。我想你可以想像得到，而你也列舉了一些例子。就可互換性而言，對測試機來說相當可互換，例如用於手機的運算引擎與用於筆電、桌機或其他任何裝置的運算引擎。比較獨特的一點是當你進入像 RF 類產品時，當然其架構會有些不同；電源管理類產品的架構也會有些不同。因此那些測試機往往會更具專用性。但我們過去也談過——我想外界看著我們會說，在手機市場有點趨於平臺期的世界裡，手機的單位成長放緩，是否預示著我們也會某種程度放慢。

And what we've always said is that, look, cell phone unit growth can slow, complexity isn't. And some of these customers are diversifying into more silicon, both in the phone and now more silicon outside the phone into compute. There are these emerging hyperscalers that are building silicon for both cloud computing and other new yet-to-be introduced consumer products. So the ability of a design team today to be formed and put together a $10 billion-plus transistor chip for a consumer application is easier and easier and easier. And as I said in my script, we work with a lot of them. Maybe a small fraction of them will be hits in the future, but they can bring entirely new classes of semiconductor -- high compute engine semiconductor applications to the market. And all of those are kind of UltraFLEX, UltraFLEXplus family devices.

而我們一直以來的說法是：你看，手機的單位成長可以放緩，但複雜度不會。而且其中一些客戶正在擴大自研矽晶的範圍，不僅在手機內，現在也把更多矽晶延伸到手機之外、進入運算領域。也有一些新興的超大規模業者（hyperscalers）正在為雲端運算以及其他尚未推出的新型消費性產品打造矽晶。因此，今天要組建一個設計團隊，並打造一顆用於消費性應用、電晶體數超過 100 億顆、價值超過 100 億美元等級的晶片，正變得越來越容易、越來越容易、越來越容易。而如我在講稿中所說，我們與其中很多團隊合作。也許只有一小部分未來會成為爆款，但它們可以把全新的半導體類別——高運算引擎的半導體應用——帶到市場上。而這些基本上都屬於 UltraFLEX、UltraFLEXplus 系列裝置。

Your next question comes from C.J. Muse of Evercore.

您的下一個問題來自 Evercore 的 C.J. Muse。

I guess, Mark, another technology question. You've kind of taught us to think about transistor count in kind of thinking about test times for mobility compute. And curious, as we go to more high-performance compute, where there are more thermal issues, perhaps more complex software algorithms, how should we be thinking about the test times in that transition?

我想，Mark，另一個技術問題。您某種程度上教我們用電晶體數量來思考行動運算的測試時間。我很好奇，當我們走向更高效能運算、面臨更多散熱議題、可能還有更複雜的軟體演算法時，我們應該如何看待在這個轉換過程中的測試時間？

Well, I think the test times have less to do with the application. The only place where the application drives really fundamentally different test times as automotive because of the issues there. It's more the technology. So I go back to C.J., what I said before about 3-nanometer is going to be more impactful to both test times in phones and in computers and in servers and in graphics and in everything else, probably than any of those end market applications.

嗯，我認為測試時間與應用本身的關係較小。唯一會因應用而在根本上驅動截然不同測試時間的地方是車用，因為那裡有相關議題。更多是技術因素。所以我回到 C.J.，就像我之前說的，3 奈米對手機、電腦、伺服器、圖形處理以及其他所有領域的測試時間影響，可能都會比任何那些終端市場應用更大。

Okay. That's helpful. And then I guess, Sanjay, a question for you. Clearly, you're making investments, particularly in IA. But as you look to 2022, can you speak to your outlook for operating leverage. And as part of that, how we should think about OpEx relative to top line growth?

好的。這很有幫助。接著我想，Sanjay，問您一個問題。很明顯你們正在投資，特別是在 IA。但當你們展望 2022 年時，能否談談你們對營運槓桿的看法？以及其中一部分，我們應該如何看待營業費用（OpEx）相對於營收成長（top line growth）的關係？

Sure. I'd say we're in the early innings of looking at 2022 and obviously, we'll provide an update in January. But OpEx is growing in kind of a couple of key areas this year, obviously tied to higher volume, we have variable compensation as well as engineering spends tied to our operations to qualify new suppliers et cetera. And then we continue to invest in both go-to-market and engineering across the test portfolio. And as you noted, we're leaning in, obviously, in -- as I noted in my prepared remarks, we're leaning into our IA investments really to help drive going forward. The other component is the G&A expenses going forward. This was a big year of OpEx growth. The only thing I'd say is that next year we're not going to be as large percentage-wise growth, but that's really all I have to say on it right now.

當然。我會說我們還在初步階段評估 2022 年，當然我們會在 1 月提供更新。但今年 OpEx 在幾個關鍵領域成長：很明顯與更高的出貨量相關，我們有變動薪酬，以及與營運相關的工程支出，用於認證新供應商等等。此外，我們也持續在整個測試產品組合上投資於市場推進（go-to-market）與工程研發。如你所提到的，我們也正在加大投入——正如我在事先準備的發言中提到的——我們加碼 IA 投資，主要是為了推動未來成長。另一個組成部分是未來的 G&A（一般及行政）費用。今年是 OpEx 成長很大的一年。我現在能說的只有：明年在百分比上不會成長得那麼多，但目前我就只能說到這裡。

Your next question comes from Timothy Arcuri, UBS.

您的下一個問題來自 UBS 的 Timothy Arcuri。

Mark, I just wanted to see if you can update us on the SOC TAM. You had said 4, 5 last call and segments of that was compute was about $1 billion. Mobility was 1.8 to 1.85. Autos were about 500, Industrial was 500, 550. I'm just kind of wondering if you can update us on those numbers.

Mark，我想看看您是否能更新一下 SOC 的 TAM。您上次電話會議提到是 4、5（十億美元）左右，其中運算（compute）約 10 億美元。行動（mobility）是 18 億到 18.5 億。車用約 5 億，工業約 5 億到 5.5 億。我只是想知道您能否更新一下這些數字。

Yes. Good question. So fundamentally, it's in the same range. It's probably trending more up towards the higher end of that range. So I think the numbers, the submarkets you got right on what we talked about last time, maybe compute is driving us a little bit higher in that range at this point in time. But it's pretty close. And memory is still at about $1 billion TAM as well.

可以。好問題。基本上仍在相同區間。可能正朝該區間的高端走。所以我認為你提到的各子市場數字，和我們上次談的差不多；也許目前是運算在推動我們略微往區間上緣走。但非常接近。而記憶體（memory）的 TAM 仍大約也是 10 億美元。

Okay. Great. And then I guess I had a question just on profitability in IA. I note, Sanjay, you just answered a question about operating leverage next year. But what's the catalyst maybe -- this time last year, we were thinking IA would be 10% to 15% op margin and then it got cut to 10% and then it went to 5%, and now it's kind of low single digits. And I get that the penetration there is very low. But what's the catalysts for you to look at maybe we shouldn't be investing so much money and maybe we can optimize OpEx investment? So I guess I'm just kind of wondering how you think about sort of what's going on at IA and what the long-term profitability target is?

好的。很好。接著我想問一個關於 IA 獲利能力的問題。我注意到，Sanjay，您剛剛回答了明年營運槓桿的問題。但觸發因素是什麼——去年這個時候，我們原本認為 IA 的營業利益率會是 10% 到 15%，後來被下修到 10%，再到 5%，現在大概是低個位數。我理解那邊的滲透率非常低。但有哪些觸發因素會讓你們覺得也許不該投入那麼多資金、也許可以最佳化 OpEx 投資？所以我想了解你們如何看待 IA 目前的狀況，以及長期的獲利目標是什麼？

Yes. So our range, as we said prior is 5% to 15%, and we expect to be low single digits, as I said in my prepared remarks. But really, how we think about it is we manage the business on a -- think about it as a rule of 40, combining the year-over-year growth with the operating profit. And fundamentally, we're -- as we see or as been noted by Mark, where penetration is very, very low and we look at all the jobs that can be automated and the scarcity of labor and economic growth, we really see a strong tailwind for the portfolio. And with that, we're very focused and conscious on both engineering and go-to-market investments. And so how we think about it is quite simple. And that is where we have a strong belief that we're going to grow and it's going to accelerate that revenue growth, we're going to lean into the investment. And again, think about it in the rule of 40. And as I said in my prepared remarks, our gross margins are actually improving in that portfolio. And so when we see that growth start to moderate, then we'll start to moderate the OpEx to get to an operating profit of, let's say, in the mid-20s.

是的。我們的區間如先前所說是 5% 到 15%，而我們預期會是低個位數，正如我在事先準備的發言中所說。但我們的思考方式是：我們以一種——可以把它想成「40 法則（rule of 40）」——來管理這個業務，把年對年成長與營業利益結合起來看。而從根本上來說——如 Mark 所見或所提到——在滲透率非常、非常低的情況下，我們看到有大量工作可以自動化，加上勞動力稀缺與經濟成長，我們確實看到這個產品組合有很強的順風。因此，我們在工程與市場推進投資上都非常聚焦且審慎。所以我們的思考其實很簡單。也就是：在我們強烈相信會成長、且營收成長會加速的地方，我們就會加大投資。同樣用 40 法則來看。而如我在事先準備的發言中所說，我們在該產品組合的毛利率其實正在改善。因此，當我們看到成長開始趨緩時，我們就會開始放緩 OpEx，讓營業利益率達到例如 20% 中段（mid-20s）。

Your next question comes from Vivek Arya, Bank of America.

您的下一個問題來自美國銀行（Bank of America）的 Vivek Arya。

Mark, this move to 3-nanometer, is that a benefit for Teradyne in '22 or '23? I thought it would be '23, but I wanted to confirm. And kind of following on for that, what are the top 1 or 2 end markets that you are most excited about in terms of growth for next year?

Mark，轉向 3 奈米，這對 Teradyne 來說是在 2022 年受益，還是 2023 年？我原本以為會是 2023 年，但想確認一下。另外延伸問一下，明年在成長方面，你們最看好、最興奮的前 1 或 2 個終端市場是哪些？

So on the 3-nanometer question, that was one of the things I mentioned that we'll be looking at carefully between now and January when we update you on the 2022 view. And we've seen some pushouts on some of those nodes for a -- by a few months. So the ability to intercept them in a meaningful way in Q -- or sorry, in 2022 is a swing factor. We don't see that exactly yet. But whatever happens in 2022, it might get some early ramping, but the bulk of it is going to be 2023, '24, '25 kind of and beyond. So a little bit could happen in '22, how much we won't know more until January.

關於 3 奈米的問題，這是我提到的其中一件事：我們會在現在到 1 月之間仔細觀察，屆時我們會更新你們對 2022 年的看法。我們也看到某些節點（nodes）有一些延後，延後了幾個月。因此，能否在 2022 年——抱歉——以有意義的方式切入，是一個變動因素（swing factor）。我們目前還看不得那麼清楚。但不論 2022 年發生什麼，可能會有一些早期爬坡（early ramping），但主要的量會在 2023、'24、'25 以及之後。所以 2022 年可能會發生一點，但會有多少，要到 1 月才會更清楚。

In terms of markets next year that are interesting and exciting, I think it goes back to these emergent hyperscalers that are developing some new applications and very complex silicon for those new applications. Some of those could latch in the market with new product introductions and drive a whole new sort of demand stream for semiconductors and the testers associated with them. So that's kind of what we're rooting for and what we're close to, and we're seeing could be breakouts for 2022.

至於明年哪些市場有趣且令人興奮，我認為要回到這些新興的超大規模業者（hyperscalers）：他們正在開發一些新應用，以及針對這些新應用所需的非常複雜的矽晶片（silicon）。其中一些可能會透過新產品導入在市場上站穩腳跟，並帶動一整股新的半導體需求流，以及與之相關的測試機需求。所以這就是我們所期待、也密切合作關注的方向；我們看到 2022 年可能會出現一些突破（breakouts）。

Got it. Very helpful. And maybe just following up on that. Is there a way, Mark, to contrast the additional complexity in a product that's going into a hyperscaler application versus a mobility application. I understand die sizes might be different, then packaging, et cetera, might be different. But in -- conceptually, what does the mix shift from a more mobility heavy end market to something that is taking you more into the compute and hyperscaler land means for Teradyne in terms of your growth prospects? And also the -- just the seasonality because mobility tends to be a lot more seasonal markets. Those other markets are perhaps less seasonal. So what does that mix shift mean for Teradyne over the long term?

了解。非常有幫助。我想再追問一下。Mark，有沒有辦法對比一下，導入超大規模雲端服務商（hyperscaler）應用的產品，相較於行動裝置（mobility）應用，會增加哪些額外的複雜度？我理解晶粒尺寸可能不同，封裝等等也可能不同。但從概念上來說，終端市場組合從較偏重行動裝置，轉向更偏向運算與 hyperscaler 領域，對 Teradyne 的成長前景意味著什麼？另外還有——季節性，因為行動裝置市場通常季節性更強。其他市場可能季節性較弱。所以，這樣的組合轉移對 Teradyne 長期而言意味著什麼？

Yes, that's a good question. And the range of devices being developed at hyperscalers is quite large. Some of them are simpler than a classic cellphone device, apps processor in a cell phone. So those aren't going to have much of an impact even if they latch. But some of them that are more, I would say, leverage AI, machine learning and high res displays are equal to or greater complexity than what you might find in a typical cell phone application for those technologies. So it's a broad spectrum is, I guess, the bottom line. But if you sort of play it out, I'd say the hyperscalers are going to, on average, bring cell phone-like complexity applications to the market. It's probably not going to be something that's -- take the $50 billion transistor device that kind of surprised everybody. It's probably not at that extreme. It's probably more in the $10 billion transistor range on average and then moving up year-over-year-over-year after that as they iterate on the design.

是的，這是個好問題。而且 hyperscaler 正在開發的裝置範圍相當廣。其中有些比經典的手機裝置——手機裡的應用處理器——更簡單。所以即使它們放量，影響也不會太大。但也有一些我會說是更偏向運用 AI、機器學習以及高解析度顯示的裝置，其複雜度與你在典型手機應用中看到的這些技術相當，甚至更高。所以總結來說，這是一個很廣的光譜。但如果把它推演下去，我會說 hyperscaler 平均而言會把類似手機等級複雜度的應用帶到市場上。大概不會是那種——像是那顆 500 億電晶體、讓大家都很驚訝的裝置。可能不會到那麼極端。平均可能更接近 100 億電晶體的範圍，之後隨著他們在設計上反覆迭代，逐年逐年往上提升。

Your next question comes from Krish Sankar with Cowen and Company.

下一個問題來自 Cowen and Company 的 Krish Sankar。

I have 2 questions. Mark, just want to ask the 3-nanometer question in a different way, you spoke about how test investments have to catch up to front end WFE spending. But during the 3D NAND investment cycle WFE grew in 2016 and '17, while the memory test spending kind of came in 2018 in a meaningful way. So I'm kind of curious, how should we think about the time lag effect of test to front end foundry logic WFE this time around?

我有兩個問題。Mark，我想用不同方式問 3 奈米的問題。你提到測試端的投資需要追上前段晶圓製造設備（WFE）的支出。但在 3D NAND 的投資循環中，WFE 在 2016 和 2017 年成長，而記憶體測試支出則是在 2018 年才以顯著方式進來。所以我有點好奇，這一次我們應該如何看待測試相對於前段晶圓代工邏輯 WFE 的時間落後效應？

Yes. Good questions. So all of the WFE investment gets recognized before single testers associated with that investment. So the way to think about test is whenever you see some silicon coming off a new node, assume that the testers were installed maybe 3 months or so prior to that. So once the 3-nanometer fab comes online and you start to see product coming off of it in reasonable volume more than sort of pilot line volume, assume that the tester installations occurred about 3 months prior. That's kind of the best guidance I can give you there. And then the only other thing I would caution about memory versus SOC is that the curve of test time to transistor count is not the same in memory as it is in SOC. Memory tends to be a bit more efficient for reasons I won't get into here. So you can double transistors and memory, and you're not going to probably move test time more than 20%, 30% or so, let's say, whereas in SOC, it's not quite linear, but it's closer.

是的。好問題。所以所有的 WFE 投資都會先被認列，之後才會認列與該投資相關的單機測試機（tester）。因此看待測試的方式是：每當你看到有一些新節點的矽晶片開始下線時，就假設測試機大概在那之前 3 個月左右就已經安裝完成。所以一旦 3 奈米晶圓廠上線，而且你開始看到有合理的出貨量——不只是試產線的量——就假設測試機的安裝大約發生在 3 個月之前。這大概是我能給你的最佳指引。另外我唯一要提醒的是，記憶體與 SoC 的情況不同：測試時間相對於電晶體數的曲線，在記憶體上與在 SoC 上並不一樣。記憶體因為一些原因（這裡就不展開）通常效率更高。所以在記憶體裡你把電晶體數翻倍，測試時間大概不會增加超過 20%、30% 左右；但在 SoC 上，雖然也不是完全線性，但更接近線性。

Got it. Got it. That's very helpful, Mark. Then I have a follow-up for Sanjay on gross margins. Despite the constrained environment, gross margins have stepped up about 200-plus basis points over the last like 4 quarters or so to a 59% to 60% range. Is this kind of the new baseline we should assume at these revenue run rates?

了解。了解。非常有幫助，Mark。接著我有一個關於毛利率的追問要問 Sanjay。儘管環境受限，過去大約 4 個季度毛利率仍提升了 200 多個基點，來到 59% 到 60% 的區間。在目前這些營收運行水準下，這是否算是我們應該假設的新基準？

Yes. I think I talked about that earlier. So we're going through our strategic planning process in Q4, and it's really the basis of our earnings model update in January. As I've said, just having some visibility into the first half of next year, I think you should expect to see that gross margin be similar in the first half of next year as we have in the second half of 2021. From -- and so there's many different variables in there. I noted that we've been managing through the component and logistics cost increase. We've had good mix and come down the cost curve of our new products and gaining operating leverage. So there's many variables in there that we're going to look at closely and as our revenue gets more and more diversified. There's a lot of puts and takes, but we'll provide more guidance on that in the January call.

是的。我想我先前有提到。我們在第四季正在進行策略規劃流程，而這其實是我們在一月更新獲利模型的基礎。如我所說，基於我們對明年上半年有一定能見度，我認為你應該預期明年上半年的毛利率會與 2021 年下半年相近。至於——其中有很多不同的變數。我提到我們一直在管理零組件與物流成本上升的影響。我們的產品組合不錯，新產品也沿著成本曲線下降，並且取得營運槓桿。所以裡面有很多變數，我們會仔細檢視；而且隨著我們的營收愈來愈多元化，會有很多相互拉扯的因素，但我們會在一月的電話會議提供更多指引。

Your next question comes from Brian Chin, Stifel.

下一個問題來自 Stifel 的 Brian Chin。

Nice results. Sorry, first to clarify something going back on the commentary. If you had more chips, you would be able to ship to an appreciably higher level of revenue in both the Semi Test and Industrial Automation businesses in Q4. Did I hear that correctly?

成果不錯。抱歉，我先回到剛才的評論釐清一點。如果你們有更多晶片，你們在第四季的半導體測試（Semi Test）與工業自動化（Industrial Automation）兩個業務，都能出貨到顯著更高的營收水準。我聽得對嗎？

In Industrial Automation, percentage-wise, we'd be able to ship more. In the Test portfolio, we think we're going to manage through the majority of the supply issues at this point.

在工業自動化方面，以比例來看，我們確實可以出貨更多。在測試產品組合方面，我們認為目前我們將能處理掉大部分的供應問題。

Okay. Okay. Great. And then interesting discussion here on 3-nanometer and sort of like a fulcrum event in terms of test intensity again at 3-nanometer or gate-all-around. Maybe a couple of questions as my follow-up. What is it about that, that -- I know it probably is a longer-form discussion, but in short form, what is it about 3-nanometer or 3-nanometer gate-all-around, is it sort of what the yield modeling suggests for kind of lower yield rates there compounded by die sizes, compounded by transistors or advanced packaging? Or is it also sort of a combination of the types of devices that you think that enables in terms of the intersection point of various customers and their kind of chip road maps? Just kind of want to get at that. And then like in kind of bigger picture, if the high-performance compute test TAM is a little bit more secular, a little bit more growthy, if you've kind of benchmarked test growth at sort of 4% to 8%, how much of a premium do you think we're looking at over the next several years in terms of HPC TAM?

好的。好的。很好。另外，這裡對 3 奈米的討論很有意思，似乎在 3 奈米或環繞式閘極（gate-all-around）上，測試強度又出現一個類似支點事件（fulcrum event）。我想追問幾個問題。這背後的原因是什麼——我知道可能需要更長篇幅，但簡短來說，3 奈米或 3 奈米的 gate-all-around 到底是什麼因素導致？是良率模型所暗示的較低良率，再加上晶粒尺寸、電晶體數或先進封裝等因素的疊加嗎？還是也包含你認為它所能啟用的裝置類型——也就是不同客戶及其晶片路線圖在某個交會點上所帶來的組合效應？我想更理解這一點。另外從更宏觀的角度，如果高效能運算（HPC）測試的 TAM 更偏向長期結構性、成長性更高，而你們對測試成長的基準大約是 4% 到 8%，那你認為未來幾年 HPC 測試 TAM 會有多大的溢價（premium）？

Okay. there's a lot in there, Brian. So let me, first of all, take on 3-nanometer. So the issues about 3-nanometer certainly enables more dense, more transistors per die. That's not what I'm talking about. That's a big driver of test in the future. That would absolutely be something that would give us a positive outlook for the next midterm. There's the chiplet thing that we talked about where you have mixed nodes going into chips, and then you need known good die testing and then the multi-packaging needs increasingly test intensity. That's not what I'm talking about. But that's real and that's there, too. The thing that I'm talking about with 3-nanometer that's unique, let's say, is the change in the transistor architecture from FinFET to gate-all-around.

好的。Brian，這裡面內容很多。所以我先談 3 奈米。關於 3 奈米的議題，確實能帶來更高密度、每顆晶粒（die）更多電晶體。但我說的不是這個。那會是未來測試需求的一大驅動因素。這絕對會讓我們對下一個中期展望更為正面。還有我們談過的 chiplet（小晶片）趨勢，也就是把不同製程節點混合整合到晶片裡，接著你需要已知良品晶粒（known good die）測試，而多晶片封裝也需要愈來愈高的測試強度。但我說的也不是這個。不過那也是真實存在的。我所說的 3 奈米獨特之處在於，例如，電晶體架構從 FinFET 轉變為環繞閘極（gate-all-around）。

That happens once per decade. Again, it happened with FinFETs earlier in the last decade. It's happening with gate-all-around here in the next few years. Those transitions, everything else being equal, introduce additional, typically, failure modes that need more test methodologies to make sure the device is functioning correctly. And so let's say, the average test time per transistor can tend to be higher because of the additional verification needed related to that new architecture. Now early in the architecture's life, that tends to be higher. As the architecture matures over time, that premium, let's say, comes down. So -- and the world doesn't shift to 3-nanometer en masse day 1 either.

這種事情大約每十年發生一次。同樣地，上一個十年早期發生的是 FinFET。接下來幾年則會發生環繞閘極（gate-all-around）。在其他條件相同的情況下，這些轉換通常會引入額外的失效模式，需要更多測試方法學來確保元件運作正確。因此可以說，平均每個電晶體的測試時間可能會因為與新架構相關的額外驗證需求而提高。而在架構生命週期的早期，這個值往往會更高。隨著架構隨時間成熟，那個溢價（premium）會下降。所以——而且世界也不會在第 1 天就大規模全面轉向 3 奈米。

So there's this bleeding up of certain devices using 3-nanometer that are highly test-intensive and more and more come online year after year, and then the learning curve comes on. But net-net, again, if you look at the test market from 2000 to 2012, it had been a declining market. And 2012 to now, it's been growing quite well in excess of 10%. Part of that is the parallel testing, ameliorating that we've talked about, part of it's the FinFET story that we've talked about and the growing transistor count. So we're at another one of those junctures with gate-all-around.

因此會有一些採用 3 奈米、測試強度很高的特定裝置先行導入，之後一年一年有更多產品上線，同時學習曲線也會發揮作用。但總體而言，如果你看 2000 到 2012 年的測試市場，它一直是下滑的市場。而從 2012 到現在，它的成長相當不錯，超過 10%。其中一部分來自我們談過的平行測試所帶來的改善（ameliorating），一部分來自我們談過的 FinFET 故事以及電晶體數量的成長。所以我們正處在另一個環繞閘極（gate-all-around）的轉折點。

And then sorry, HPC and...

然後抱歉，HPC 和……

Test HPC higher or lower...

HPC 的測試是更高還是更低……

Okay. Yes. So the -- the growth rate in compute is certainly going to be at the higher end of our mix of submarkets, we think, over the midterm. So if you think the average market growth rate, we're going to update this in January, but pick a number, is 8%. We think compute is probably leading that by at least a couple of points.

好的。是的。所以我們認為，在中期來看，運算（compute）的成長率肯定會位於我們各個子市場組合的較高端。所以如果你把整體市場平均成長率——我們會在一月更新——先隨便取個數字，例如 8%。我們認為運算（compute）可能至少會領先它幾個百分點。

And operator, we have time for one more question, please.

接線員，我們還有時間再問最後一個問題，謝謝。

Your final question comes from Sidney Ho, Deutsche Bank.

最後一個問題來自德意志銀行（Deutsche Bank）的 Sidney Ho。

So I have 2 quick ones. The first one is, you talk about supply constraints impacting the IA business. I'm just curious, the gross margin has been pretty good overall, you talk about offering the leverage is one of the reasons. To the extent that your input cost on logistics and freight cost increase, both in your test and IA business, are you able to pass along some of those costs to your customers?

我有兩個簡短問題。第一個是，你提到供應限制正在影響 IA 業務。我只是好奇，整體毛利率一直相當不錯，你提到提供槓桿（leverage）是原因之一。如果你的物流與運輸成本等投入成本上升，無論在測試業務或 IA 業務，你們能否把其中一部分成本轉嫁給客戶？

So at this point, we've been managing through it. So materially, no.

所以就目前而言，我們一直在設法管理並度過這些狀況。所以在實質上，不能。

Okay. Maybe my follow-up question is, a lot of discussion on 3 nanometers today, but it's really an opportunity in 2023 and beyond. But if you kind of look at next year, I know you're going to update us in January, how does this impact your tester business, if a customer chooses to move from instead of going to 3 nanometers, they go to a different generation of 5 nanometers? What is the tester reuse rate when you compare the 2 nodes between the 3 nanometers and the alternative?

好的。我的追問是，今天對 3 奈米有很多討論，但那其實是 2023 年及之後的機會。但如果你看明年，我知道你們會在一月更新我們，這會如何影響你們的測試機（tester）業務：如果客戶選擇不走 3 奈米，而是改走不同世代的 5 奈米呢？在比較 3 奈米與替代方案之間的兩個節點時，測試機的重複使用率（reuse rate）是多少？

Yes. The tester reuse, people will reuse 5-nanometer generation testers in the 3-nanometer era. So a new tester is not required. And testers tend to have a useful life of a decade or more at a customer, and they can span usability across many, many nodes and many, many generations of devices. So there's nothing in terms of a new tester here that comes with 3-nanometer. And yes, by and large, it's a 2023 and beyond type story. But absent that story, what we've been seeing for the past 10 years in test, since 3-nanometer in 2022 is likely not going to be a huge event because it's likely to be later in the year, people will still move down the complexity curve at 5- and 5-plus nanometer nodes and increase transistor counts and such along that path. So even absent 3-nanometer, we have an optimistic view of what '22 looks like.

是的。測試機重複使用方面，人們會在 3 奈米時代重複使用 5 奈米世代的測試機。所以不需要新的測試機。而測試機在客戶端的使用壽命通常可達十年或更久，並且可以跨越非常多的節點與非常多世代的裝置持續使用。所以 3 奈米並不會帶來「必須新增一台新測試機」這件事。而且是的，大致上這是 2023 年及之後的故事。但即便沒有那個故事，我們在過去 10 年於測試市場看到的情況仍然成立；由於 2022 年的 3 奈米很可能不會是重大事件，因為它可能會在年內較晚才發生，人們仍會在 5 奈米與 5 奈米加（5-plus）節點沿著複雜度曲線往下走，並在那條路徑上增加電晶體數量等等。所以即便沒有 3 奈米，我們對 2022 年的展望仍然樂觀。

We'll probably start out the year similar to how we started out years in the past and at the beginning of the year with some modest growth in the first quarter and then it kind of swings on what's going to happen over the summer, which is our peak quarters, around the refreshes. Will those refreshes of silicon be in 3-nanometer, will they be in 5 plus with the traditional normal transistor count growth that we've seen for the past 10 years, those are things that we have to get closer to next year to really understand.

我們可能會以類似過去幾年年初的方式開始新的一年：第一季有一些溫和成長，接著就要看夏季會發生什麼——那是我們的旺季季度，圍繞著產品更新（refresh）。這些矽晶（silicon）的更新會是在 3 奈米嗎？會是在 5 奈米加，並延續我們過去 10 年看到的傳統、正常的電晶體數量成長嗎？這些都是我們必須更接近明年才能真正理解的事情。

Okay, everybody. We are out of time. Thanks so much for joining us today. And those in the queue, I'll get back to you later today. And again, thanks, everyone, for joining. Bye-bye.

好的，各位。我們時間到了。非常感謝大家今天加入。在等候隊列中的各位，我今天稍晚會再回覆你們。再次感謝各位的參與。再見。

Ladies and gentlemen, this concludes today's conference call. Thank you participating. You may now disconnect.

各位女士先生，今天的電話會議到此結束。感謝各位參與。您現在可以掛線。