A 16 nm 1.60TOPS/W High Utilization DNN Accelerator with 3D Spatial Data Reuse and Efficient Shared Memory Access
By Xiaoling Yi, Ryan Antonio, Yunhao Deng, Fanchen Kong, Joren Dumoulin, Jun Yin, Marian Verhelst
MICAS-ESAT, KU Leuven, Leuven, Belgium
Abstract
Achieving high compute utilization across a wide range of AI workloads is crucial for the efficiency of versatile DNN accelerators. This paper presents the Voltra chip and its utilization-optimised DNN accelerator architecture, which leverages 3-Dimensional (3D) spatial data reuse along with efficient and flexible shared memory access. The 3D spatial dataflow enables balanced spatial data reuse across three dimensions, improving spatial utilization by up to 2.0x compared to a conventional 2D design. Inside the shared memory access architecture, Voltra incorporates flexible data streamers that enable mixed-grained hardware data pre-fetching and dynamic memory allocation, further improving the temporal utilization by 2.12-2.94x and achieving 1.15-2.36x total latency speedup compared with the non-prefetching and separated memory architecture, respectively. Fabricated in 16nm technology, our chip achieves 1.60 TOPS/W peak system energy efficiency and 1.25 TOPS/mm2 system area efficiency, which is competitive with state-of-the-art solutions while achieving high utilization across diverse workloads.
Index Terms — DNN Accelerator, 3D Spatial Data Reuse, Flexible and Efficient Data Access, Shared Memory, High Utilization
To read the full article, click here
Related Semiconductor IP
- ReRAM NVM in DB HiTek 130nm BCD
- UFS 5.0 Host Controller IP
- PDM Receiver/PDM-to-PCM Converter
- Voltage and Temperature Sensor with integrated ADC - GlobalFoundries® 22FDX®
- 8MHz / 40MHz Pierce Oscillator - X-FAB XT018-0.18µm
Related Articles
- 3+ ways to design reconfigurable algorithm accelerator in IP block
- RAID6 accelerator in a PowerPC IOP SOC
- Emulator, accelerator, prototype - what’s the difference?
- NVMe host IP for computing accelerator
Latest Articles
- An FPGA-Based SoC Architecture with a RISC-V Controller for Energy-Efficient Temporal-Coding Spiking Neural Networks
- Enabling RISC-V Vector Code Generation in MLIR through Custom xDSL Lowerings
- A Scalable Open-Source QEC System with Sub-Microsecond Decoding-Feedback Latency
- SNAP-V: A RISC-V SoC with Configurable Neuromorphic Acceleration for Small-Scale Spiking Neural Networks
- An FPGA Implementation of Displacement Vector Search for Intra Pattern Copy in JPEG XS