On October 19th, Beijing Mingxin Qirui Technology Co., Ltd. (hereinafter referred to as Mingxin Qirui) announced the completion of a Series A financing round of over 400 million yuan, led by Boyuan Capital, with follow-on investments from Bosch Venture Capital, MiHoYo, GigaDevice, and other investors. Existing shareholders such as Xiaomi Strategic Investment also continued to increase their stakes. The company stated that the funds from this round will be used for the industrialization and commercialization of RRAM (resistive memory) as well as the research and development of the new product LPU (language processing unit). Relevant reports did not disclose the valuation or equity dilution ratio for this round of financing.
On the same day, two other developments were announced. Lu Tao, the former president of Graphcore and general manager of Greater China, joined the company as co-founder of CEO. Lu Tao has previously served as the general manager of Cavium in China and vice president of Xizhi Technology, with experience in server CPU, AI processors, and the field of silicon photonics. The company also launched the development of the first generation of LPU, aimed at large model inference scenarios, with a focus on optimizing Token generation speed, inference latency, and service costs.
This change was preceded by technical preparations. On October 14th, Mingxin Qirui announced the mass production information of its first-generation 1Gb RRAM chip, known as Mount Tai No.1. The company stated that Mount Tai No.1 is based on the 1TnR technology route (where one selection transistor corresponds to multiple resistive memory cells), specifically a 1T4R structure, which means one selection transistor corresponds to four resistive memory cells, and it uses a 28-nanometer CMOS manufacturing process. The company disclosed a storage density of 30Mb/ mm², an access latency of less than 20 nanoseconds, and it integrates on-chip hardware error correction circuits.
At present, Taishan-1 has completed tape-out and is in the process of verification; no progress regarding mass production has been disclosed yet.
Mingxin Qirui was established in May 2024, with core team members coming from the Chinese Academy of Sciences and Fudan University. The team initially focused on RRAM research and development as well as industrialization, and is now expanding into large model inference chips.
In terms of product layout, Mingxin Qirui stated that the company currently has three main product lines.
The first product is the embedded RRAM IP, aimed at display drivers, microcontrollers, SSD controllers, and security chips. The company states that it has already entered customer scenarios and plans to expand its scale, but the names of customers, order amounts, or shipment volumes have not been disclosed. The second product is the high-capacity RRAM, also known as Taishan No.1, which is still in the verification phase. The third product is a hybrid computing solution that combines RRAM with DRAM, targeting embodied intelligence and AI PC type edge-side inference scenarios. The company is exploring mixed 3D integration, but specific chip models, tape-out progress, or commercialization timelines have not been revealed.
Lu Tao stated externally that the first generation of LPU will not adopt RRAM; instead, a more mature storage medium will be used. The product's goal is to support models with trillions of parameters and achieve 1000 Token per user per second. This metric represents a research and development goal, and there is currently no actual performance data to support it.
Mingxin Qirui confirmed with All-Weather Technology that LPU is currently in the research and development phase, and no further timeline for tape-out, verification, or delivery has been disclosed.
From the disclosed information, Mingxin Qirui's technical layout presents three characteristics, although its performance and cost advantages remain to be verified. First, the company has extended from storage devices to inference systems, initiating research and development on both RRAM IP and standalone memory simultaneously, aiming to master storage media, computing architecture, and system synergy capabilities at the same time. This approach is not unique to Mingxin Qirui; other companies in the industry, such as Xinyi Technology, already cover RRAM storage and AI inference chips. Second, 1TnR is a verifiable technical design. The company claims that circuit solutions such as negative voltage design, multi-voltage domain wordline drive, and on-chip hardware error correction can improve leakage, miswriting, and read reliability, but there is a lack of third-party test data and comparative studies with competitors to substantiate these advantages. Third, the LPU roadmap shares similar goals with Groq. As of now, Mingxin Qirui has not made the complete architecture public, so a rigorous comparison of chip architectures is not possible.
From the perspective of industrialization progress, some companies within the industry have disclosed advancements in mass production or commercialization. According to the official website of Xinyi Technology, its standalone memory chips have been mass-produced since 2021, with a cumulative shipment of over 30 million units; Yizhu Technology claims to have mass-produced all-digital integrated storage and computing chips based on new storage technologies; Houmo Intelligence has launched integrated storage and computing chips such as M30, M50, and AI. International manufacturer Groq has already provided commercial inference services based on LPU through GroqCloud. In contrast, Mingxin Qirui's Taishan No.1 is still in the verification phase, and its first-generation LPU is in the early stages of development, so the timing of product delivery and commercialization scale remains to be seen.
Mingxin Qirui is currently expanding its inference system from RRAM storage to AI. In the future, it will still be necessary to focus on whether these technical capabilities can be transformed into mass production orders and customer revenue.
Material consistency, durability, yield, process compatibility, and software ecosystem are issues that typically need to be addressed during the industrialization of new types of storage. It is worth further observing whether the verification of Taishan No.1 has been completed, the progress of customer sample submissions, whether the specific names of customers and the scale of orders for embedded RRAM IP can be disclosed, and the timeline for the first generation of LPU to enter mass production.
