Tower Semiconductor Begins Mass Production of AI Communication Components
Tower Semiconductor, in collaboration with Israeli firm NewPhotonics, has announced the commencement of mass production for integrated laser optical engines designed for data centers. These components are manufactured on Tower's silicon photonics platform, integrating lasers, optical amplifiers, modulators, and detectors onto a single chip.
The newly produced components support data transfer rates of 800 gigabits to 1.6 terabits per second. The companies are planning a more significant advancement, aiming to introduce components capable of 6.4 terabits per second in the first half of 2027. These next-generation components are intended for AI system architectures, including Scale-Up and Scale-Out.
Optical engines are crucial for converting electrical information into optical signals and vice versa. As the number of accelerators in data centers increases, the bottleneck shifts to the interconnecting network, necessitating higher data transfer speeds and lower power consumption. NewPhotonics' products are designed for both pluggable communication modules and Near-Chip Optics (NPO) solutions, which place optical components closer to processing chips to minimize signal loss and power usage.
Doron Tal, Senior VP and General Manager at NewPhotonics, stated that the next phase of AI infrastructure demands optical connectivity that can scale in both production volume and performance. He highlighted that the synergy between NewPhotonics' chip design and Tower's manufacturing platform enabled the transition to production for equipment manufacturers and data centers.
This announcement reinforces silicon photonics as a key growth driver for Tower, particularly in the AI infrastructure sector. The company has previously announced collaborations with Nvidia on 1.6T optical modules and with Marvell, having shipped over 5 million integrated photonic chips. Tower also announced in July an expansion of its manufacturing capacity in Japan for silicon photonics and SiGe, including a move to 300mm wafer production, with a business model targeting $3.6 billion in revenue and $1.2 billion in net profit by 2028, explicitly allocating investments to meet AI and data center demand.
