Cube architecture targets powerful edge AI devices, says Winbond
Affordable edge AI computing can be enabled in mainstream use cases with the CUBE customised ultra-bandwidth elements) architecture which has been introduced by Winbond Electronics. The architecture optimises memory technology for seamless performance running generative AI on hybrid edge/cloud applications.
CUBE enhances the performance of front-end 3D structures such as chip on wafer (CoW) and wafer on wafer (WoW), as well as back-end 2.5D/3D chip on Si-interposer on substrate and fan-out solutions, said Winbond. It is compatible with memory density from 256Mbit to 8Gbit with a single die, and it can also be 3D stacked to enhance bandwidth while reducing data transfer power consumption.
Winbond said CUBE is a major step forward, enabling seamless deployment across various platforms and interfaces. The technology is suited to advanced applications such as wearable and edge server devices, surveillance equipment, ADAS and co-robots, said the company.
The company believes that the integration of cloud AI and powerful edge AI will define the next phase of AI development. “With CUBE, we are unlocking new possibilities and paving the way for improved memory performance and cost optimisation on powerful edge AI devices,” added the company.
CUBE is power efficient, consuming less than 1pJ/bit, ensuring extended operation and optimised energy usage. It has bandwidth capabilities ranging from 32Gbytes per second to 256Gbytes per second per die for accelerated performance that exceeds industry standards.
Memory capacities range from 256Mbit to 8Gbit per die, based on the 20nm specification now and 16nm in 2025. This allows CUBE to fit into smaller form factors seamlessly. The introduction of through-silicon vias (TSVs) further enhances performance, improves signal and power integrity while reducing the I/O area through a smaller pad pitch, said Winbond. It also reduces heat dissipation, especially when using SoC on the top die and CUBE on the bottom die, advised the company.
It is also cost-effective, said Winbond, boosting the data rate up to 2Gbits per second with total 1K I/O. When paired with legacy foundry processes like 28nm/22nm SoC, CUBE unleashes high bandwidth capabilities, reaching 32 to 256Gbytes per second, equivalent to harnessing the power of 4-32pcs LP-DDR4x 4266Mbits per second x16 I/O bandwidth.
Stacking the SoC (top die without TSV) on top of the CUBE (bottom die with TSV) minimises the SoC die size, eliminating any TSV penalty area, said Winbond.
The company said CUBE can unleash the full potential of hybrid edge/cloud AI to elevate system capabilities, response time, and energy efficiency.
The company added that it is actively engaging with partner companies to establish the 3DCaaS platform, which will leverage CUBE’s capabilities.
Winbond Electronics supplies semiconductor memory products with capabilities of product design, R&D, manufacturing and sales services. The company’s product portfolio consists of specialty DRAM, mobile DRAM, code storage flash and TrustME secure flash for tier-1 customers in communication, consumer electronics, automotive and industrial, and computer peripheral markets.
Winbond is headquartered in Central Taiwan Science Park (CTSP), and has subsidiaries in the USA, Japan, Israel, China, Hong Kong and Germany and 12 inch fabs in Taichung and Kaohsiung in Taiwan.
