ReRAM-based Processor-In-Memory (PIM) architectures have demonstrated their ability to accelerate the matrix multiplication performed by the Transformer. However, these approaches often shift the performance bottleneck from the attention mechanism to the softmax computation. Moreover, data sharding, commonly employed for acceleration, hinters the Transformer's ability to find global softmax value. This paper proposes SegTransformer, a ReRAM-based PIM accelerator that improves softmax performance with high accuracy through segmentation techniques. Experimental results indicate that the SegTransformer significantly outperforms state-of-the-art Transformer accelerators.

Deterministic databases provide strong serializability while avoiding concurrency-control related aborts by establishing a serial ordering of transactions before their execution. Recent work has shown that they can also handle skewed and contended workloads effectively. These properties are achieved by batching transactions in epochs and then executing the transactions within an epoch concurrently and deterministically. However, the predetermined serial ordering of transactions makes these databases more vulnerable to long-latency transactions. As a result, they have mainly been designed as main-memory databases, which limits the size of the datasets that can be supported.