TensorPool: A 3D-Stacked 8.4TFLOPS/4.3W Many-Core Domain-Specific Processor for AI-Native Radio Access Networks

TL;DR

TensorPool is a specialized many-core processor designed for AI-native radio access networks, achieving high tensor computation throughput and energy efficiency within strict power and latency constraints.

Contribution

We introduce TensorPool, a domain-specific, 3D-stacked many-core processor with tensor acceleration, optimized for AI-based 6G RAN, demonstrating significant performance and efficiency improvements.

Findings

TensorPool achieves 3643 MACs/cycle with 89% tensor-unit utilization.

It provides 6× more tensor performance than a core-only cluster.

TensorPool improves GOPS/W/mm² efficiency by 9.1×.

Abstract

The upcoming integration of AI in the physical layer (PHY) of 6G radio access networks (RAN) will enable a higher quality of service in challenging transmission scenarios. However, deeply optimized AI-Native PHY models impose higher computational complexity compared to conventional baseband, challenging deployment under the sub-msec real-time constraints typical of modern PHYs. Additionally, following the extension to terahertz carriers, the upcoming densification of 6G cell-sites further limits the power consumption of base stations, constraining the budget available for compute ($\leq$ 100W). The desired flexibility to ensure long term sustainability and the imperative energy-efficiency gains on the high-throughput tensor computations dominating AI-Native PHYs can be achieved by domain-specialization of many-core programmable baseband processors. Following the domain-specialization strategy, we present TensorPool, a cluster of 256 RISCV32IMAF programmable cores, accelerated by 16 256 MACs/cycle (FP16) tensor engines with low-latency access to 4MiB of L1 scratchpad for maximal data-reuse. Implemented in TSMC's N7, TensorPool achieves 3643~MACs/cycle (89% tensor-unit utilization) on tensor operations for AI-RAN, 6$\times$ more than a core-only cluster without tensor acceleration, while simultaneously improving GOPS/W/mm$^2$ efficiency by 9.1$\times$. Further, we show that 3D-stacking the computing blocks of TensorPool to better unfold the tensor engines to L1-memory routing provides 2.32$\times$ footprint improvement with no frequency degradation, compared to a 2D implementation.