Samsung in urgent need of boosting foundry yield 삼성, AI 프로세서 파운드리 수율 향상시급

[김우정]

반도체 기업은 설계를 전문으  팹리스, 칩리스기업과 생산전문 파운드리, 종합반도체 기업 IDM 등이 있습니다.

대표적인 팹리스기업 엔비디아의 젠슨황 CEO는 AI프로세서의 제조를 TSMC에서 변경할 것을 암시 했습니다.

대만의 TSMC와 함께 대표적인 파운드리 기업인 삼성전자는 수율문제로 공급을 못하고 있습니다.

삼성전자가 수율증대를 통해 3분기에 5조5,000억원 감소한 반도체 사업의 영업이익이 회복되기를 기대합니다.

Samsung in urgent need of boosting foundry yield for AI processors
삼성, AI 프로세서를 위한 파운드리 수율 향상이 시급

By Nam Hyun-woo 남현우 기자
Posted : 2024-09-12 16:57 Updated : 2024-09-12 17:00
Korea Times

Nvidia CEO Jensen Huang speaks at SIGGRAPH 2024 in Denver, July 29. AP-Yonhap 엔비디아 CEO 젠슨 황이 7월 29일 덴버에서 열린 SIGGRAPH 2024에서 연설하고 있다. AP-연합

Nvidia CEO hints at switching suppliers from TSMC
엔비디아 CEO, TSMC에서 공급업체 변경암시

Samsung Electronics faces an urgent need to enhance its foundry manufacturing yield for artificial intelligence (AI) processors as Nvidia has suggested the possibility of moving orders away from Taiwan’s TSMC, citing sustained strong demand for its chipsets.
삼성전자는 엔비디아가 칩셋에 대한 지속적인 강력한 수요를 이유로 대만 TSMC에서 주문을 옮길 가능성을 시사함에 따라 인공지능(AI) 프로세서의 파운드리 제조 수율을 강화해야 할 시급한 상황에 직면해 있다.

During a Goldman Sachs technology conference in San Francisco, Wednesday (local time), Nvidia CEO Jensen Huang told the audience that TSMC’s “agility and their capability to respond to our needs is just incredible,” but “if necessary, of course, we can always bring up others.”
수요일(현지 시간) 샌프란시스코에서 열린 골드만삭스 기술 컨퍼런스에서 엔비디아의 CEO 젠슨 황은 청중들에게 TSMC의 "민첩성과 우리의 요구에 대응하는 능력은 놀라울 뿐,"이지만 "물론 필요하다면 언제든지 다른 업체를 언급할 수 있다."고 말했다.

Shrugging off concerns about AI chip demand, Huang said demand for Nvidia's Blackwell chips is "so great," and that Nvidia’s suppliers are doing their best to keep up with it.
젠슨 황은 AI 칩 수요에 대한 우려를 일축하며 엔비디아의 블랙웰 칩에 대한 수요가 "매우 크다,"며 엔비디아의 공급업체들이 이를 따라잡기 위해 최선을 다하고 있다고 말했다.

At the same time, he noted that Nvidia could switch suppliers if necessary, due to concerns like rising geopolitical tensions in Taiwan. However, he warned that such a change might "result in lower-quality products."
동시에 그는 대만의 지정학적 긴장 고조와 같은 우려로 인해 필요한 경우 공급업체를 전환할 수 있다고 언급했다. 그러나 그는 이러한 변화가 "품질이 낮은 제품으로 이어질 수 있다."고 경고했다.

Nvidia is outsourcing the physical production of most of its advanced AI processors to the world's largest foundry, TSMC, rather than to other foundries, including the second-largest, Samsung Electronics.
엔비디아는 대부분의 첨단 AI 프로세서의 실물 생산을 두 번째로 큰 삼성전자를 포함한 다른 파운드리가 아닌 세계 최대 파운드리인 TSMC에 아웃소싱하고 있다.

Last year, Samsung’s foundry secured Nvidia’s orders for 8-nanometer-based chips like the Tegra for automobiles and the RTX-3000 graphics processing unit (GPU) in 2020. However, it has not been successful in securing orders for more advanced AI processors, such as the H series or Blackwell, which require more advanced manufacturing processes.
작년에 삼성 파운드리는 2020년에 자동차용 테그라, RTX-3000 그래픽 처리 장치(GPU)와 같은 8나노미터 기반 칩을 엔비디아로부터 수주했다. 그러나 고급 제조 공정이 필요한 H 시리즈나 블랙웰과 같은 고급 AI 프로세서의 수주는 성공적으로 이루어지지 못했다.

Samsung Electronics' Pyeongtaek Campus in Gyeonggi Province / Courtesy of Samsung Electronics 삼성전자 경기도 평택캠퍼스 / 삼성전자 제공

The question, as Huang noted, is the quality. Samsung's foundry is scrambling to improve its manufacturing yield, which refers to the ratio of good and error-free units to the total number of units produced. But the number is not improving as fast as Samsung hopes.
젠슨 황이 언급했듯이 문제는 품질이다. 삼성 파운드리는 총 생산량 대비 우수한 유닛과 오류 없는 유닛의 비율을 의미하는 제조수율을 개선하기 위해 분주히 움직이고 있다. 하지만 그 수는 삼성이 기대하는 것만큼 빠르게 개선되지 않고 있다.

Samsung's foundry struggled with a 3-nanometer process yield that remained in the single digits until the first quarter of this year, resulting in delays in supplying engineering samples for its own Exynos 2500 chipsets.
삼성의 파운드리는 올해 1분기까지 한 자릿수를 유지한 3나노미터 공정 수율로 인해 어려움을 겪었고, 이로 인해 자체 엑시노스 2500 칩셋의 엔지니어링 샘플 공급이 지연되었다.

Domestic analysts assume that Samsung managed to improve the yield to nearly 20 percent in the second quarter, but this is still insufficient for mass production, which requires a yield of at least 60 percent.
국내 분석가들은 삼성이 2분기에 수익률을 20% 가까이 개선한 것으로 추정하지만, 최소 60% 이상의 수익률을 요구하는 대량생산에는 아직 부족한 수준이다.

Against this backdrop, Samsung appears to be struggling to secure foundry orders. According to industry officials, the company has recently adjusted its plans for installing equipment in its newest fab, P4 in Pyeongtaek, Gyeonggi Province, to prioritize the production of advanced DRAM memory, such as high-bandwidth memory (HBM) chips, first.
이러한 배경에서 삼성은 파운드리 수주 확보에 어려움을 겪고 있는 것으로 보입니다. 업계 관계자에 따르면 삼성은 최근 경기도 평택에 위치한 최신 팹인 P4에 장비를 설치할 계획을 고대역폭 메모리(HBM) 칩과 같은 첨단 DRAM 메모리 생산에 우선순위를 두기로 조정했다.

The fab was initially scheduled to start with equipment installation for NAND, followed by foundry and then DRAM products. However, this order has been altered due to sluggish foundry orders. There are rumors that Samsung may decide to use P4 exclusively for manufacturing memory chips, given the stable demand for high-bandwidth memory (HBM) and other advanced memory products used in AI servers.
이 팹은 처음에 낸드용 장비 설치로 시작하여 파운드리와 DRAM 제품으로 이어질 예정이었다. 그러나 파운드리 주문이 부진하면서 주문이 변경되었다. 삼성이 고대역폭 메모리(HBM) 및 AI 서버에 사용되는 기타 고급 메모리 제품에 대한 안정적인 수요를 고려하여 메모리 칩 제조에 P4를 독점적으로 사용하기로 결정할 수 있다는 소문이 돌고 있다.

In this context, Samsung’s investment in its plant in Taylor, Texas, is facing questions. The company initially planned to use the plant to mass produce 4-nanometer chips starting next year, but has postponed this plan until 2026. While the yield for the 4-nanometer process is reportedly stable, Samsung is struggling to secure orders from fabless companies.
이러한 맥락에서 삼성의 텍사스 테일러 공장 투자는 의문에 직면해 있다. 삼성은 당초 내년부터 이 공장을 사용하여 4나노미터 칩을 대량 생산할 계획이었지만, 이 계획을 2026년까지 연기했다. 4나노미터 공정의 수율은 안정적인 것으로 알려졌지만 삼성은 팹리스 기업으로부터 주문을 확보하는 데 어려움을 겪고 있다.

As a result, there is also speculation that Samsung might shift its focus to 2-nanometer chips instead of 4-nanometer ones to secure orders for more advanced products in the future. However, reports indicate that the company is also facing challenges in improving the manufacturing yield for the 2-nanometer process, as well as the 3-nanometer process.
이에 따라 삼성이 향후 고급 제품 주문을 확보하기 위해 4나노 칩 대신 2나노 칩으로 초점을 전환할 것이라는 추측도 나오고 있다. 그러나 보고서에 따르면 삼성은 2나노 공정과 3나노 공정의 제조수율을 개선하는 데도 어려움을 겪고 있는 것으로 나타났다.

“One of the core reasons of Samsung foundry’s cumulative losses is the low yield,” a semiconductor industry official said. “The company managed to stabilize its yield for the 4-nanometer process, but has not done that yet in more advanced processes, such as second generation 3-nanometer or 2-nanometer processes.”
반도체 업계 관계자는 "삼성 파운드리의 누적 손실의 핵심 원인 중 하나는 낮은 수율 때문"이라고 말했다. "삼성 파운드리는 4나노미터 공정의 수율을 안정화하는 데 성공했지만 2세대 3나노미터 또는 2나노미터 공정과 같은 고급 공정에서는 아직 그렇게 하지 못했다."

Samsung does not disclose the separate earnings of its foundry business, but market observers estimate that the company has incurred approximately 1.5 trillion won ($1.12 billion) in operating losses during the first half of this year.
삼성은 파운드리 사업의 별도 실적을 공개하지 않지만, 시장 전문가들은 올해 상반기 동안 약 1조 5,000억 원(11억 2,000만 달러)의 영업 손실을 입은 것으로 추정하고 있다.

“Samsung’s achievements in the HBM business still falls short of what should be expected from a brand like Samsung, and its foundry division is still struggling with losses,” Eugene Investment & Securities analyst Lee Seung-woo said. “As a result, the operating profit for Samsung’s semiconductor business in the third quarter of this year is expected to decline to 5.5 trillion won.”
유진투자증권 이승우 애널리스트는 "삼성의 HBM 사업 성과는 여전히 삼성과 같은 브랜드에서 기대해야 할 수준에 미치지 못하고 있으며 파운드리 사업부도 여전히 손실에 시달리고 있다,"며 "이에 따라 올해 3분기 삼성 반도체 사업의 영업이익은 5조 5,000억 원으로 감소할 것으로 예상된다."고 전망했다.

#Foundry파운드리 #Yield수율 #AIprocessors인공지능프로세서 #Nvidia엔비디아 #TSMCTSMC(대만 반도체 제조업체) #Chips칩 #DRAMDRAM #HBM(High-BandwidthMemory)고대역폭메모리 #3-nanometerprocess3나노미터공정 #Operatinglosses영업손실