Energy-efficient and fast memristor-based serial multipliers applicable in image processing
Computation-in-memory (CIM) is a promising technique for overcoming the Von-Neumann bottleneck. Applying memristors in CIM reduces data transfer between processor and memory. Memristive CIM reduces energy consumption and processing time of data-intensive applications. Multipliers are one of the arit...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Results in Engineering |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S259012302500101X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Computation-in-memory (CIM) is a promising technique for overcoming the Von-Neumann bottleneck. Applying memristors in CIM reduces data transfer between processor and memory. Memristive CIM reduces energy consumption and processing time of data-intensive applications. Multipliers are one of the arithmetic circuits that play a significant role in data-intensive processing applications. Serial Material Implication (IMPLY) logic design implements arithmetic circuits by applying emerging memristive technology that enables CIM Array (CIM-A). The computational complexity of IMPLY-based multipliers for use in the CIM-A architecture is a significant design challenge. Implementing IMPLY-based crossbar array-friendly multipliers and reducing their computational cycles and energy consumption are designers' goals in applying computational applications such as convolution in the basic CIM-A architecture. This work presents unsigned and signed array multipliers using serial IMPLY logic. The proposed multipliers have improved significantly compared to State-Of-the Art (SOA) by applying the proposed Partial Product Units (PPUs) and overlapping computational steps. The number of computational steps, energy consumption, and required memristors of the proposed 8-bit unsigned array multiplier are improved by up to 36 %, 31 %, and 47 % compared to the classic designs. The proposed 8-bit signed multiplier has also improved the computational steps, energy consumption, and required memristors by up to 59 %, 54 %, and 45 %. The performance of the proposed multipliers in the applications of Gaussian blur and edge detection is also investigated, and the simulation results have shown an improvement of 31 % in energy consumption and 33 % in the number of computational steps in these applications. |
|---|---|
| ISSN: | 2590-1230 |