HSCO-Bench is the first end-to-end benchmark for LLM agents performing hardware-software co-design of heterogeneous SoCs, where only two of five frontier models produced valid FPGA prototypes that underutilized available hardware resources.
1.1 Computing’s energy prob- lem (and what we can do about it)
7 Pith papers cite this work. Polarity classification is still indexing.
7
Pith papers citing it
representative citing papers
A 194M-parameter spiking dual-path model trained on 3B Chinese-English tokens achieves held-out PPL 8.88-8.93 at >89% per-element sparsity, trailing GPT-2 201M by 7.7% while showing that LIF temporal integration outperforms simple top-k masking at matched sparsity.
Replacing pointwise convolutions with DWHT yields a model with 79.1% fewer parameters, 48.4% fewer FLOPs, and 1.49% higher accuracy than MobileNet-V1 on CIFAR-100.
Nonlinear detuning stabilizes non-adiabatic magnonic dynamics in YIG:Co nanostructures, enabling low-occupancy resonant states with estimated 22 aJ switching energy.
KLR Hopfield networks exhibit robustness to quantization but sensitivity to pruning, interpreted as arising from dense bimodal parameterization of sparse input mappings.
citing papers explorer
-
HSCO-Bench: An Agent-Driven End-to-End Hardware-Software Co-design Benchmark for Systems-on-Chip
HSCO-Bench is the first end-to-end benchmark for LLM agents performing hardware-software co-design of heterogeneous SoCs, where only two of five frontier models produced valid FPGA prototypes that underutilized available hardware resources.
-
SymbolicLight V1: Spike-Gated Dual-Path Language Modeling with High Activation Sparsity and Sub-Billion-Scale Pre-Training Evidence
A 194M-parameter spiking dual-path model trained on 3B Chinese-English tokens achieves held-out PPL 8.88-8.93 at >89% per-element sparsity, trailing GPT-2 201M by 7.7% while showing that LIF temporal integration outperforms simple top-k masking at matched sparsity.
-
New pointwise convolution in Deep Neural Networks through Extremely Fast and Non Parametric Transforms
Replacing pointwise convolutions with DWHT yields a model with 79.1% fewer parameters, 48.4% fewer FLOPs, and 1.49% higher accuracy than MobileNet-V1 on CIFAR-100.
-
Nonlinear Stabilization of Non-Adiabatic Magnonic Dynamics
Nonlinear detuning stabilizes non-adiabatic magnonic dynamics in YIG:Co nanostructures, enabling low-occupancy resonant states with estimated 22 aJ switching energy.
-
Quantization robustness from dense representations of sparse functions in high-capacity kernel associative memory
KLR Hopfield networks exhibit robustness to quantization but sensitivity to pruning, interpreted as arising from dense bimodal parameterization of sparse input mappings.
- Function, Complexity and Thermodynamics in Adaptive and Intelligent Soft Matter Systems: An Information-Theoretical Formulation
- Hardware-Software Co-Design of Scalable, Energy-Efficient Analog Recurrent Computations