Xiaomi and Huawei develop low-latency RAM for smartphones, optimizing artificial intelligence performance
Large Chinese companies, including Huawei and Xiaomi, are at the forefront of developing innovative RAM technology for mobile devices. The new architecture, known as Low Latency Wide DRAM (LLW), seeks to overcome the main challenge of current memories: the high latency in transmitting crucial information to the processor, especially for executing artificial intelligence tasks locally.
The creation of LLW is strongly inspired by the principles of efficient High Bandwidth Memory (HBM). These modules, notorious for their extreme speeds, are widely used in data centers. Its distinguishing feature lies in its ability to vertically stack chips in 3D arrangements close to the processor, facilitating a massive connection to the CPU for ultra-fast data transmission.
On the same topic: Demand for AI and lack of RAM drive price rises in GPUs and SSDs
The concept behind LLW aims to adapt this advanced HBM approach to the smartphone environment. The intention is to significantly expand the communication channels and position the memory even closer to the processor. This is done through a direct integration design that also helps mitigate overheating issues.
The main innovation of Low Latency Wide lies in its ability to not depend on a single path to communicate with the processor. Just like HBM, the strong point of this new technology is that it employs multiple smaller communication lines, which allow data to be sent in a more balanced and, consequently, extremely fast manner.
The urgency to improve the RAM memory standard
In a global scenario of component shortages, the initiative to develop a new category of memory for cell phones could seem risky. However, the evolution of LLW is intrinsically linked to the growing demand for artificial intelligence, aiming to enable smartphones to run AI models directly on the device, reducing or eliminating dependence on cloud services.
More on this story: Global increase in memory prices forces smartphone manufacturers to rescue drop notch and slow screens
See Also
Anonymous fan funds Thimbleweed Park 2 and frees creator from publishers
Consortium led by Jeff Bezos and Eduardo Saverin buys 30% of Liverpool
Cristiano Ronaldo and Georgina’s agreement shields fortunes in official union
The big challenge for engineers is the impossibility of simply transferring HBM memories to cell phones. This is mainly due to the limited internal space in these devices, which contrasts drastically with the large area available in AI servers, capable of accommodating large volumes of memory around a processor.
An additional difficulty associated with this technology is thermal management. Despite its high speed, HBM memory requires sophisticated cooling systems, which are unfeasible for the size of cell phones. For this reason, LLW adopts the data sending concept of HBM, but without replicating the vertical stacking of chips.
The industry’s attraction to this new generation of RAM lies in the significant reduction in latency. This implies that the memory can exchange data with the processor more quickly and in greater quantities, translating into a substantial performance gain for the system when executing complex operations.
Learn more: Experts project announcement of the iPhone 18 Pro line for August 2026
According to tipster Fixed Focus Digital, LLW promises a reduction in energy consumption of up to 50%, in addition to a 1.5x performance increase compared to the LPDDR5X standard. The expectation is that more details about the debut of this technology, which still seems a little distant, will be released from the second half of 2027.
The relevance of the memory sector is evidenced by recent data: the global DRAM industry reached revenues of more than R$500 billion in the first quarter of this year, marking record revenue. This significant amount highlights the economic weight and the constant search for technological innovations in this highly competitive and essential market for the evolution of electronic devices.
