Event
Dosilicon, a Chinese memory-chip designer, disclosed its “memory-compute fusion” plan: it intends to invest about RMB 680 million of overfunded proceeds in a “memory-compute fusion chip R&D project” to build low-power embedded smart-computing chips for on-device AI, and it hosted 87 institutional investors on October 30 to discuss niche-memory pricing and its on-device AI roadmap.
Technology / Product Details
Dosilicon is one of the few domestic companies that can offer NAND, NOR, DRAM, and MCP solutions at once. Its memory-compute fusion chip integrates memory, connectivity, and computing into a single die through advanced packaging, targeting smart home, industrial IoT, consumer electronics, and healthcare on-device AI scenarios. Once complete, it will create synergy with its memory main business, turning a “memory die” into a “memory-in-compute” compute unit and cutting the power and footprint of on-device AI.
From the Industry Landscape…
On-device AI is opening new curves not only for compute chips but also for memory vendors. Niche memory faces tightened supply and rising prices as overseas fabs shift capacity to higher-margin products, so memory makers are extending toward “memory + compute,” shifting from “supplying die” to “supplying smart units.” For embedded and edge devices, the fusion of storage and compute will significantly reshape terminal functional density and cost structure.
Junyi Business Link
The fusion of storage and compute is also reshaping edge-board design thinking. Junyi Microelectronics’ core boards J-3568 / J-3576 / J-3588 and edge motherboards in the K series pair compute and storage for edge AI, edge computing hosts 5G-EDGE-3399Pro / AI-BOX-3588 serve edge nodes needing local models and data caching, and the industrial mini PC IPC-i5-1235U and the all-in-one touch PC complete the terminal forms for smart-home and industrial-IoT products.
Source: Dosilicon: low-power embedded smart chip for on-device AI (2026-11-03)