Caffeine: Towards uniformed representation and acceleration for deep convolutional neural networks
By Chen Zhang, Zhenman Fang, Peipei Zhou, Peichen Pan, Jason Cong. 2016 IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2016
357 citations recorded by the source corpus.
Subfields: Pharmacology, Toxicology
View Paper: https://www.semanticscholar.org/paper/e982fc566b0b98e94bf95b224e2b9c743eb503cc
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Pharmacology
In Toxicology
Citations in the corpus
- A Scalable Near-Memory Architecture for Training Deep Neural Networks on Large In-Memory Datasets (2018, background): “FPGA-based accelerators report energy efficiencies on the order of 10 Gop/s W and usually rely on fixed-point arithmetic and less than 32 bit precision [32], [33].”
- Fast inference of deep neural networks in FPGAs for particle physics (2018, background): “Caffeine [29] is another CNN accelerator for Caffe-specified models targeting Xilinx devices that support a SDAccel15 environment and a PCIe interface between the FPGA and a host.”
- Understanding Reuse, Performance, and Hardware Cost of DNN Dataflow: A Data-Centric Approach (2018, methodology): “Caffeine [48] proposed a full automatic FPGA flow that includes pragma-based annotation in programs, dataflow optimization framework, and DSE for FPGAs based on the analytic model defined over loop tiling and unrolling.”
- Escher: A CNN Accelerator with Flexible Buffering to Minimize Off-Chip Transfer (2017, methodology): “The weight-major method in [11] requires the batch size to match the number of vector dot product units in the accelerator, also limiting the design to small batch sizes.”
- FINN-R (2018, unknown): “] AlexNet AWSF1 230 1,682 545,000 1,884.2(-) - - W8A16 Wei etal. [61] VGG16 AWSF1 240 1,690 504,000 2,260.9(-) - - W8A16 Wei etal. [61] ResNet AWSF1 230 1,644 496,000 1,875.3(-) - - W8A16 Zhang etal. [67] VGG16 VC709 150 - - 354(-) 26 14 W16A16 Aydonat etal. [6] AlexNet GX1150 303 - - 1,382(-) ∼45 3”
- Recent Advances in Convolutional Neural Network Acceleration (2018, background): “Caffeine [127] Xilinx KU060 FPGA 200MHz 365 GOPS”
- FPGA-Based Accelerators of Deep Learning Networks for Learning and Classification: A Review (2019, methodology): “[153], [162] analyzed the transformation of CONV and FC layers to regular matrix multiplication presented in prior work [98].”
- Gist: Efficient Data Encoding for Deep Neural Network Training (2018, methodology): “While using FPGAs [50] and ASICs [28], [25] is also possible, most of these designs are either proprietary or in relatively early development stages.”
- Automated systolic array architecture synthesis for high throughput CNN inference on FPGAs (2017, methodology): “The design in [10] adopts a systolic-like architecture to mitigate the timing issue for the large design, but it still directly connects all PEs to the on-chip memory and results in not fully local interconnects.”
- A survey of FPGA-based accelerators for convolutional neural networks (2018, methodology): “Similarly, the BW of FPGA is much smaller than that of GPU ( * 20 versus 700GB/s) [7, 12–15].”
- FP-DNN: An Automated Framework for Mapping Deep Neural Networks onto FPGAs with RTL-HLS Hybrid Templates (2017, methodology): “[27] designed uniformed covolution kernel for both convolutional layers and fullyconnected layers.”
- Angel-Eye: A Complete Design Flow for Mapping CNN Onto Embedded FPGA (2018, methodology): “Zhang et al. [11] and Sharma et al. [12] implemented automatic design tools but are targeting a single network.”
- AccPar: Tensor Partitioning for Heterogeneous Deep Learning Accelerators (2020, unknown): “…[33], Neural Cache [123,124], Diffy [125] FPGA FPGA-DCNN [34], Embedded-FPGA-CNN [35], FPGA-Exploration [36], OpenCL-FPGA-CNN [37] [126], Caffeine [38], DeepBurning [39], FPGA-DPCNN [40], TABLA [41], DNNWEAVER [42], FP-DNN [43], FPGA-LSTM [44], ESE [45], FPGA-Dataflow [46], FPGA-BNN [47],…”