Microprocessors are used in a variety of systems ranging from high-performance super
computers running scientific applications to battery powered cell phones performing realtime
tasks. Due to the large disparity between processor clock speed and main memory
access time, most modern processors include several caches, which consume more than half
of the total chip area and power budget. As the performance gap between processors and
memory has increased, the trend has been to increase the size of the on-chip caches.
However, increasing the cache size also increases its access time and energy consumptions.
This growing power dissipation problem is making traditional cooling and packaging
techniques less effective thus requiring cache designers to focus more on architectural level
energy efficiency than performance alone.
The goal of this thesis is to propose a new cache architecture and to evaluate its
efficiency in terms of miss rate, system performance, energy consumption, and area
overhead. The proposed architecture employs the use of a few Ternary-CAM (TCAM)
cells in the tag array to enable dynamic compression of tag entries containing contiguous
values. By dynamically compressing tag entries, the number of entries in the tag array can
be reduced by 2N, where N is the number of tag bits that can be compressed. The architecture described in this thesis is applicable to any cache structure that uses Content
Addressable Memory (CAM) cells to store tag bits.
To evaluate the effectiveness of the TCAM Enhanced Cache Architecture for a wide
scope of applications, two case studies were performed ?? the L2 Data-TLB (DTLB) of a
high-performance processor and the L1 instruction and data caches of a low-power
embedded processor. Results indicate that a L2 DTLB implementing 3-bit tag compression
can achieve 93% of the performance of a conventional L2 DTLB of the same size while
reducing the on-chip energy consumption by 74% and the total area by 50%. Similarly, an
embedded processor cache implementing 2-bit tag compression achieves 99% of the
performance of a conventional cache while reducing the on-chip energy consumption by
33% and the total area by 10%.
| Identifer | oai:union.ndltd.org:tamu.edu/oai:repository.tamu.edu:1969.1/2314 |
| Date | 29 August 2005 |
| Creators | Surprise, Jason Mathew |
| Contributors | Mahapatra, Rabi N. |
| Publisher | Texas A&M University |
| Source Sets | Texas A and M University |
| Language | en_US |
| Detected Language | English |
| Type | Book, Thesis, Electronic Thesis, text |
| Format | 240334 bytes, electronic, application/pdf, born digital |
Page generated in 0.0039 seconds