논문 상세보기
pp.529-536
Semiconductors, optimized for artificial intelligence (AI) applications, are efficiently handling large-scale data processing and complex computations with high speed and low power consumption. They accelerate AI model training and inference in data centers, cloud services, autonomous vehicles, and mobile devices. As demand for high-speed data transmission and extensive data processing grows, global companies are developing proprietary AI semiconductors, and subsequently, high-density packaging technologies are needed to interconnect multiple processor chips. To achieve this, an interposer is required. An interposer is a layer used in packaging technology for combining multiple chips, which includes wiring that is inserted to electrically connect a semiconductor chip with a substrate that has a significant pitch difference. Among the materials employed as substrates or interposers, organic, silicon and glass are being considered. While silicon interposers are usually used to connect the main substrate and multiple chips, producing very thin silicon wafers and controlling warpage is challenging, and so they suffer from poor yield and integration. Also, organic substrates have difficulty achieving fine pitch because of their uneven surface and warpage. On the other hand, glass substrates and interposers have good electrical and thermal properties. For this reason, this study investigated AI semiconductor packaging trends and through glass via (TGV) technology, emphasizing the importance of suitable glass material selection, reliable glass-metal bonding and application to solder bumping on TGV. Advances in AI and TGV technologies are expected to drive next-generation AI semiconductor packaging development.
1. 서 론
2. 인공지능 반도체
2.1. 신경처리장치(neural processing unit, NPU)
2.2. 엣지 컴퓨팅(edge computing)
2.3. 뉴로모픽 반도체
3. 유리 코어 기판과 인터포저
3.1. 유리 소재의 특성과 인터포저의 적합성
4. AI 반도체 패키징
4.1. 칩렛(chiplet)
4.2. TGV 기술과 솔더링
4.3. 유리 기판 상 금속 코팅 및 TGV 범핑 적용
5. 결 론
Acknowledgement
References
