Wei Shaojun ends up making cores! 14nm surpasses 4nm. Why does Oriental computing core?

On July 13, 2026, the "Oriental Computing Core Software-Defined Computing Power Chip, System and Roadmap Conference" with the theme "Oriental Paradigm Leads the New Trend of AI Computing Power" was held at the Dongjiao Hotel in Shanghai. As an AI chip company established just over two years ago, Oriental Computing Core officially released the first AI chip - DF1000 - based on its new "software-defined chip" + 3D stacked "near memory computing" technology route, opening up an independent innovation path for domestic high-end computing chips that does not rely on advanced processes and HBM. The three major constraints on domestic AI computing power chips are advanced processes, high-performance memory, and supply chains. In recent years, the advancement of semiconductor process technology has slowed down significantly, transistor sizes are gradually approaching physical limits, and Moore's Law is becoming ineffective. The traditional path of relying on process shrinkage to improve chip performance is bringing diminishing returns. At the same time, the R&D and manufacturing costs required to pursue advanced processes have increased exponentially, making the traditional "upgraded process" model of driving performance growth unsustainable. Especially in the context of the continuous explosion of artificial intelligence (AI) demand for computing power, process dividends can no longer provide a performance jump that is sufficient to cover these needs, resulting in an increasingly strong sense of tear between "computing power hunger" and "cost gap". On the other hand, the core feature of the von Neumann architecture still used in current mainstream computer systems is that the computing unit and the storage unit are separated from each other, and data needs to be frequently transferred between the two - reading from the memory before calculation and writing back after calculation. This "separation of storage and calculation" design was acceptable in early general computing, but in the context of the explosive growth of data-intensive scenarios such as artificial intelligence, big data analysis, and high-performance computing, shortcomings have become increasingly prominent. Typical problems include: power wall (the energy consumption of data transfer is much higher than the calculation itself), performance wall (the computing speed of the processor far exceeds the memory bandwidth, and the computing unit is often in a state of "waiting for data"), and the memory wall (the growth of storage capacity and bandwidth lags far behind the computing power demand). The three superimpose each other and seriously restrict the overall performance of the system. Currently, as AI computing power demand has begun to shift from training to inference applications, Decode still occupies the current