As genomes, transcriptomes and meta-genomes are being sequenced at a faster pace than ever, there is a pressing need for efficient genome assembly methods. Two practical issues in assembly are heavy memory usage and long execution time during the read indexing phase. In this article, a parallel and memory-efficient method is proposed for reads indexing prior to assembly. Specifically, a hash-based structure that stores a reduced amount of read information is designed. Erroneous entries are filtered on the fly during index construction. A prototype implementation has been designed and applied to actual Illumina short reads. Benchmark evaluation shows that this indexing method requires significantly less memory than those from popular assemblers.