Winbond Launches CUBE-Lite: A New Semi-Custom Memory Solution for Edge AI

In a significant advance for Edge AI technology, Winbond Electronics Corporation has announced the introduction of CUBE-Lite, a semi-custom memory solution that promises high bandwidth coupled with low power consumption. This innovation caters to a diverse range of applications including AI-ISP (Image Signal Processor), wearables, and microcontroller (MCU) based systems among various embedded AI designs.

The demand for efficient memory solutions is skyrocketing as AI processing shifts towards endpoint devices. Here, microcontrollers, microprocessors, and neural processing units are in constant need of greater memory bandwidth to handle intensive AI workloads, all while facing the critical challenge of maintaining low power consumption. CUBE-Lite has been engineered specifically for power-sensitive Edge AI applications, boasting impressive capabilities that reach up to 8GB/s in memory bandwidth while consuming approximately 70mW. This results in energy efficiency on the order of 1 picojoule per bit, crucial for applications that require both speed and responsiveness.

One of the standout features of CUBE-Lite is its simplified interface architecture. This revolutionary design provides an innovative alternative for memory integration within Edge AI System on Chips (SoCs). Winbond's engineering data suggests that this advancement can reduce power consumption in SoC controllers by over ten times when compared to traditional LPDDR4x implementations, leading to significant battery savings in mobile and portable devices. Furthermore, the simplified interface can also streamline the associated Physical (PHY) die area and integration demands for SoCs, granting developers more flexibility in achieving the necessary bandwidth for AI processing.

CUBE-Lite operates with a voltage configuration that includes VDD1 at 1.8V, VDD2 at 1.1V, and VDDQ spanning from 0.9V down to 0.75V. Such support for low operating voltages makes it an excellent match for low-power AI SoCs.

According to a spokesperson from Winbond, “CUBE-Lite extends the advantages of our acclaimed CUBE technology into a broader array of Edge AI applications. This semi-custom approach significantly lowers the entry barriers for customers seeking high memory bandwidth and extremely low power consumption, all while being finely tuned to meet the specific demands of applications such as AI-ISP, wearables, and microcontroller devices.”

The primary targets for CUBE-Lite will include AI-ISP applications, wearables, and MCUs, as well as systems focused on low-power edge vision and embedded AI that utilize Micro Linux and various AI models. Currently, CUBE-Lite is in its product validation phase, with mass production slated for the first half of 2027.

About Winbond
Founded in September 1987 and listed on the Taiwan Stock Exchange in 1995, Winbond Electronics Corporation has firmly established itself as a leading global provider of semiconductor memory solutions. Headquartered in the Central Taiwan Science Park (CTSP) in Taichung, Taiwan, the company operates two highly automated and intelligent 12-inch wafer fabs, one in Taichung and the other in Kaohsiung. Winbond's dedication to offering comprehensive integrated circuit solutions to customers worldwide encompasses vital product lines like Code Storage Flash and Customized Memory Solutions, among many others.

Winbond's continuous innovation and commitment to quality assurance help it maintain a strong reputation in the memory semiconductor industry, making it a key player in providing efficient and powerful memory solutions for the next generation of technology.