Data Processing and Information Classification-An In-Memory Approach.

Area of Science:

• Computer Engineering
• Data Science
• Embedded Systems

Background:

• The proliferation of sensors generates massive data, necessitating efficient processing.
• Remote data processing in server farms leads to significant battery drain.
• The Memory Wall problem limits performance due to slow memory access.

Purpose of the Study:

• To introduce a Processing-In-Memory (PIM) hardware accelerator.
• To demonstrate PIM's effectiveness for local data classification using Bitmap Indexing.
• To address the limitations of traditional microprocessors and memory architectures.

Main Methods:

• Design and synthesis of a reconfigurable PIM hardware accelerator using CMOS technology.
• Implementation of the Bitmap Indexing algorithm on the PIM architecture.
• Evaluation of the system's performance and power consumption.

Main Results:

• The PIM system successfully processes and classifies large datasets locally.
• The proposed architecture achieves very low power consumption.
• The system demonstrates reconfigurability for diverse tasks and standard memory operation.

Conclusions:

• Processing-In-Memory is a viable solution to overcome the Memory Wall problem.
• Local data processing with PIM significantly reduces energy consumption.
• PIM architectures offer efficient and flexible solutions for data-intensive applications.