Sebastian Siegel

Nathan Leroux, Paul-Philipp Manea, Chirag Sudarshan et al.

Transformer networks, driven by self-attention, are central to Large Language Models. In generative Transformers, self-attention uses cache memory to store token projections, avoiding recomputation at each time step. However, GPU-stored projections must be loaded into SRAM for each new generation step, causing latency and energy bottlenecks. We present a custom self-attention in-memory computing architecture based on emerging charge-based memories called gain cells, which can be efficiently written to store new tokens during sequence generation and enable parallel analog dot-product computation required for self-attention. However, the analog gain cell circuits introduce non-idealities and constraints preventing the direct mapping of pre-trained models. To circumvent this problem, we design an initialization algorithm achieving text processing performance comparable to GPT-2 without training from scratch. Our architecture respectively reduces attention latency and energy consumption by up to two and five orders of magnitude compared to GPUs, marking a significant step toward ultra-fast, low-power generative Transformers.

Sebastian Siegel, Ming-Jay Yang, John-Paul Strachan

Processing long temporal sequences is a key challenge in deep learning. In recent years, Transformers have become state-of-the-art for this task, but suffer from excessive memory requirements due to the need to explicitly store the sequences. To address this issue, structured state-space sequential (S4) models recently emerged, offering a fixed memory state while still enabling the processing of very long sequence contexts. The recurrent linear update of the state in these models makes them highly efficient on modern graphics processing units (GPU) by unrolling the recurrence into a convolution. However, this approach demands significant memory and massively parallel computation, which is only available on the latest GPUs. In this work, we aim to bring the power of S4 models to edge hardware by significantly reducing the size and computational demand of an S4D model through quantization-aware training, even achieving ternary weights for a simple real-world task. To this end, we extend conventional quantization-aware training to tailor it for analog in-memory compute hardware. We then demonstrate the deployment of recurrent S4D kernels on memrisitve crossbar arrays, enabling their computation in an in-memory compute fashion. To our knowledge, this is the first implementation of S4 kernels on in-memory compute hardware.

Sebastian Siegel, Ming-Jay Yang, Younes Bouhadjar et al.

Structured State Space models (SSM) have recently emerged as a new class of deep learning models, particularly well-suited for processing long sequences. Their constant memory footprint, in contrast to the linearly scaling memory demands of Transformers, makes them attractive candidates for deployment on resource-constrained edge-computing devices. While recent works have explored the effect of quantization-aware training (QAT) on SSMs, they typically do not address its implications for specialized edge hardware, for example, analog in-memory computing (AIMC) chips. In this work, we demonstrate that QAT can significantly reduce the complexity of SSMs by up to two orders of magnitude across various performance metrics. We analyze the relation between model size and numerical precision, and show that QAT enhances robustness to analog noise and enables structural pruning. Finally, we integrate these techniques to deploy SSMs on a memristive analog in-memory computing substrate and highlight the resulting benefits in terms of computational efficiency.