Small block architecture

Large-scale integrated circuits (ICs) face significant challenges due to non-ideal effects that arise as the size of the IC increases. One critical issue is yield loss, which grows exponentially with the area of the chip, making manufacturing large ICs costly and inefficient. Additionally, RC delay, which increases with the square of the distance, can severely limit the speed of data transfer across the chip. Other factors, such as leakage current and parasitic capacitance, which are proportional to length, also significantly contribute to inefficiencies in power consumption and performance.

​

The small block architecture addresses these size-induced limitations by dividing a large IC into thousands or even millions of smaller blocks. The design uses a novel signal transfer architecture to synchronize and operate these small blocks in parallel, enabling the performance and functionality of a large-scale IC without the aforementioned size-induced problems. This approach not only achieves extremely high performance but also operates at remarkably low power.

​

Small block architecture has been successfully proven on silicon and has multiple backing patents. It is a foundational solution for logic-compatible embedded DRAM and an ideal architecture for large-scale AI processors where efficiency, performance, and scalability are critical. AI models continue to exponentially grow in size and power consumption, putting extreme strain on existing systems and power grids. Small block architecture enables unmatched parallel processing and eliminates memory bottlenecks, enabling today's and tomorrow's largest AI models to run even faster while consuming a thousands of times less power.