Geo News
Community curated by people like you
LatestAICryptoHealthWorld AffairsUS Politics
Semicon Taiwan 2026 highlights AI chip interconnect challenges and co-packaged optics
00

Semicon Taiwan 2026 highlights AI chip interconnect challenges and co-packaged optics

Aug 30, 2026

SEMICON Taiwan 2026 launched in Taipei with record-setting scale, hosting over 1,300 exhibitors and 100,000 attendees. The event highlights a critical shift in AI hardware engineering: as conventional copper wires hit physical limits, the industry is transitioning to silicon photonics and co-packaged optics (CPO) to resolve data movement bottlenecks. TSMC's COUPE platform has entered mass production, while Samsung unveiled its 'CUBE' 3D memory roadmap to bypass physical board space constraints.

SEMICON Taiwan 2026 event scale

  • ▪SEMICON Taiwan 2026 opened its forum program on August 31, 2026 in Taipei, with the main exhibition scheduled for September 2 to 4, 2026 at TaiNEX Halls 1 and 2.
  • ▪SEMI's April 2026 300mm Fab Outlook projects global 300mm fab equipment spending will reach $133 billion in 2026 and $151 billion in 2027, crossing the $150 billion threshold for the first time.
  • ▪SEMICON Taiwan 2026 set record scales with over 1,300 exhibitors, 4,300 booths, 18 national pavilions, and an expected attendance of more than 100,000 professionals from 65 countries.
  • ▪Global semiconductor manufacturing equipment sales reached $135.1 billion, representing a 15% year-over-year increase, according to SEMI data cited at SEMICON Taiwan 2026.

AI chip interconnect bottleneck

  • ▪Conventional copper interconnects are reaching physical limits in bandwidth, power consumption, and transmission distance at AI-cluster speeds exceeding 100 terabits per second per node.
  • ▪Frontier AI training compute is growing four- to fivefold annually, pushing power constraints and data movement to the center of the semiconductor performance race.
  • ▪Data movement in modern AI systems consumes more energy than computation itself, shifting the primary engineering bottleneck from individual chip performance to system architecture.

Silicon photonics deployment

  • ▪AUO Group's Daxin Materials subsidiary showcased a system-level Micro LED co-packaged optics module integrating Ennostar transmitters and Tyntek photodetector receivers for short-reach AI data center interconnects.
  • ▪The first standalone Silicon Photonics Global Summit in SEMICON Taiwan history opened on August 31, 2026, co-chaired by TSMC's K.C. Hsu and ASE Inc.'s Dr. C.P. Hung.
  • ▪Silicon photonics uses optical signals instead of electrical signals to integrate photonic and electronic integrated circuits, reducing power consumption and data transfer costs.

Co-packaged optics production

  • ▪Nvidia partnered with TSMC on co-packaged optics and began shipping next-generation Spectrum-X switches using the technology, with production ramping up in the second half of 2026.
  • ▪Powertech Technology, partnering with Broadcom, plans low-volume production for optical engine components in 2026 before integrating them into full co-packaged optics switches in 2027.
  • ▪Co-packaged optics integrates optical engines directly within the switch ASIC package, reducing the signal path from centimeters over copper traces to millimeters over optical connections.
  • ▪United Microelectronics Corp completed its first mass-produced silicon photonics wafers at its Singapore fab in partnership with Silith Technology.
  • ▪Hon Hai Precision Industry expects to begin co-packaged optics switch shipments in Q3 2026, while its affiliate ShunSin Technology achieved mass production capability for 51.2-Tbps and 102.4-Tbps products.

TSMC COUPE platform

  • ▪TSMC identified wafer testing, fiber array units, and high-speed optical packaging assembly as the three key breakthrough areas determining whether its COUPE platform scales to volume.
  • ▪TSMC's Compact Universal Photonic Engine (COUPE) platform and 'COUPE on Substrate' co-packaged optics solutions entered mass production in 2026.

Samsung CUBE memory strategy

  • ▪Samsung is targeting 2028 for sampling its zNAND-O flash solution, which aims for 10 times the bit density of DRAM and seven times the read bandwidth of conventional NAND.
  • ▪Samsung is developing eighth-generation HBM, designated HBM5, targeting twice the performance of HBM4E, 20% higher performance per watt, and 20% lower thermal resistance.
  • ▪Samsung Electronics unveiled its 'CUBE' memory strategy at the Memory Executive Summit on September 1, 2026, focusing on vertically integrated 3D memory structures to increase capacity.
  • ▪Samsung is developing zHBM, aiming for eight times the performance of HBM4E and three times the performance per watt, while reducing thermal resistance by 75% to 90%.

3 sources

Techtimes
SEMICON Taiwan 2026 Kicks Off: AI Chips' Bottleneck Is Wires Connecting Them
View source article
Taipeitimes
Semicon to put spotlight on CPO, silicon photonics - Taipei Times
View source article
Koreaherald
Samsung unveils memory road map at Semicon Taiwan
View source article

Featured stories

View more in Semiconductor manufacturing

TSMC affiliate Vanguard considers second Singapore chip plant

Sep 28, 2026 · 2 sources

OpenAI and Synopsys partner to develop AI model for chip design

Sep 30, 2026 · 3 sources

Bank of England warns AI debt surge poses market correction risk

Sep 30, 2026 · 5 sources

DeepSeek releases software tools for Huawei AI chips to challenge Nvidia

Sep 30, 2026 · 4 sources

Story comments

Loading comments…

Related entities

TaipeiTaiwan

Related Projects

Samsung

Topics

Semiconductor manufacturingAISilicon photonicsAI hardwareData center infrastructureCompute, chips & AI infrastructure

Featured stories

View more in Semiconductor manufacturing

TSMC affiliate Vanguard considers second Singapore chip plant

Sep 28, 2026 · 2 sources

OpenAI and Synopsys partner to develop AI model for chip design

Sep 30, 2026 · 3 sources

Bank of England warns AI debt surge poses market correction risk

Sep 30, 2026 · 5 sources

DeepSeek releases software tools for Huawei AI chips to challenge Nvidia

Sep 30, 2026 · 4 sources