CN

What is Computing-In-Memory (CIM)?

Download white paper

The next-generation computing technology

CIM achieves the complete integration of the storage unit and the computing unit. The computing operations are accomplished by the storage units inside the memory chip, achieving an order of magnitude improvement in computing energy efficiency.


True integration of memory and computing

No separate compute units. Operations occur within memory.

Near-Memory Computing

•  Concept: Decouples storage and computation but places data closer to computing units.

•  Advantages: Reduces data movement latency and power consumption.

•  Limitations: 

Physical bottlenecks: Performance gains are marginal and costly with current mainstream chip designs.

Processing in memory

Tightly Coupled Memory-Compute Integration

• Memory and computing units are co-located at the architectural level

• Utilizes dedicated processing elements embedded within memory arrays

Computing in memory (CIM)

Directly eliminate the boundaries between "storage" and "computing", without the need to move data back and forth.
It significantly enhances the performance of chips, greatly reduces power consumption and keeps costs under control, making it a true integration of memory and computing.

Mature Process Compatibility

COST 70%
Delivers comparable compute power at 70% lower cost
vs von Neumann architectures
CONSUMPTION 90%
Cuts 80–90% power consumption
or equivalent compute tasks
Process 7 Years
Matches traditional architectures' performance
using 7-year-old fabrication tech
PERFORMANCE 4~10X
4–10× faster
in real-world tests(with unoptimized algorithms)