Defect-Tolerant Nanocomputing Using Bloom Filters

Abstract

We propose a novel defect-tolerant design methodology using Bloom filters for defect mapping for nanoscale computing devices. It is a general approach that can be used for any permanent defects incurred during the manufacturing process. Our redundant design methodology does not rely on a voting strategy, thus it utilizes the device redundancy more effectively than existing approaches. Additionally, our method does not have false-positive in defect identification, i.e. it will not report a defective device as functional. Moreover, it is very space economic and can be programmed to fit different scales and characteristics of the underlying specific nanoscale devices used in the system.

Combining static and dynamic defect-tolerance techniques for nanoscale memory systems

“…Tahoori [17] proposes a defect unaware design flow which identifies universal defect free subsets within the partially defective chips and reduces the size of the required defect map. Wang et al [19] proposes the use of Bloom filters for storing defect maps for nanoscale devices. Hashing for every bit, however, is expensive computationally and may significantly increase the memory access times.…”

“…While the use of Bloom filters for identifying faulty nanoscale bits [19] is promising, the scheme at the bit-level does not produce high yield in memory as the probability of finding a set of contiguous bits becomes pretty low with increasing bit error rate. More importantly, we concentrate on improving the reliability of memory blocks that will allow memory systems to continue to operate at the level of pages which can be as large as 4 KB each.…”

“…This compares to 60% of the total memory needed by bit-level Bloom filters as described in Reference [19]. Raising the granularity of Bloom filters, however, means that the lower-level bits must be protected dynamically through error correction, and this comes with its own additional cost.…”

“…A number of defect-tolerant design methods have been proposed [4][16] [19] to deal with high error rates. These methods either use redundancy such as Error Correcting Codes (ECC) [10], or reconfiguration in post manufacturing process to map out the defective regions.…”

“…Therefore, a more compact way of storing the reconfiguration data is needed. Wang et al [19] proposed the use of a Bloom filter [1] for storing the defect map. As the bit error rate increases, however, the Bloom filter will need a large amount of reliable memory to store the bit-level defect map.…”