Block Floating Point (BFP) is a type of quantization that combines high dynamic range with low-cost inference. BFP can be implemented efficiently on FPGA hardware and, at low precision, halves the logic footprint versus blocked FP16 while maintaining accuracy. Moving to very low precision halves the logic footprint again and retraining allows the recovery of any accuracy lost in transition. This paper describes our approach to achieving target accuracy and FPGA resource usage in a low-precision end-to-end AI solution. We go on to investigate the effects of retraining with our software model that replicates the low-level implementation of BFP on FPGA. Our solution allows efficacy testing for the quantization of custom networks and provides accuracy indications and resource usage for the final application. Using our solution, we were able to quantize ResNet 50, SSD300 and UNet to int5/4bfp precision without losing accuracy while reducing FPGA resources and improving performance.

This paper describes architectural enhancements in Intel® Agilex™ FPGAs and SoCs. Agilex devices are built on Intel's 10nm process and feature next-generation programmable fabric, tightly coupled with a quad-core ARM processor subsystem, a secure device manager, IO and memory interfaces, and multiple companion transceiver tile choices. The Agilex fabric features multiple logic block enhancements that significantly improve propagation delays and integrate more effectively with the second-generation HyperFlexAgilex™ pipelined routing architecture. Routing connections are re-designed to be point-to-point, dropping intermediate connections featured in prior FPGA generations and replacing them with a wider variety of shorter wire types. Fine-grain programmable clock skew and time-borrowing were introduced throughout the fabric to augment the slack-balancing capabilities of HyperFlex registers. DSP capabilities are also extended to natively support new INT9/BFLOAT16/FP16 formats. Together, along with process and circuit enhancements, these changes support more than 40% performance improvement over the Stratix® 10 family of FPGAs.

This paper describes architectural enhancements in the Altera Stratix? 10 HyperFlex? FPGA architecture, fabricated in the Intel 14nm FinFET process. Stratix 10 includes ubiquitous flip-flops in the routing to enable a high degree of pipelining. In contrast to the earlier architectural exploration of pipelining in pass-transistor based architectures, the direct drive routing fabric in Stratix-style FPGAs enables an extremely low-cost pipeline register. The presence of ubiquitous flip-flops simplifies circuit retiming and improves performance. The availability of predictable retiming affects all stages of the cluster, place and route flow. Ubiquitous flip-flops require a low-cost clock network with sufficient flexibility to enable pipelining of dozens of clock domains. Different cost/performance tradeoffs in a pipelined fabric and use of a 14nm process, lead to other modifications to the routing fabric and the logic element. User modification of the design enables even higher performance, averaging 2.3X faster in a small set of designs.

This paper describes architectural enhancements in the Altera Stratix-V" FPGA architecture, built on a 28nm TSMC process, together with the data supporting those choices. Among the key features are time borrowing flip-flops, a doubling of the number of flip-flops per LUT compared to previous Stratix architectures, a simplified embedded 20kb dual-port RAM block, and error correction that can correct up to 8 adjacent errors. Arithmetic performance is significantly improved using a fast adder with two levels of multi-bit skip. We also describe how the routing architecture and layout is optimized for the 28nm process to take advantage of a wider range of wire thicknesses offered on the different layers, and improvements in performance and routability are obtained without dramatic changes to the repeated floorplan of the logic plus routing fabric.

This paper describes architectural enhancements in the Stratix-III" and Stratix-IV" FPGA architectures. These architectures feature programmable power management, which allows the power and performance of logic and routing to be varied to minimize total power without any performance loss. This paper describes the technique used for programmable power management, and describes the experimental evaluation that led to the choice of regions in these architectures. The memory architecture is also explored by adding heterogeneous memory mapping to the FPGA Modeling Toolkit, and used to explore LUT based memory structures. The ALM structure provides more inputs than required for a simple 6 LUT, which can be used with simple modifications to efficiently support simple dual-ported LUT based RAM. Replacing the Stratix-II" small memory blocks with LUT RAM and changing the size of other two memories is shown to reduce overall core area across a set of benchmark designs.