Optimize
Optimize models for the highest performance with sparsity regularization and quantization aware training. Fine-tune existing models or train from scratch.
Hardware
Introducing SPU-001
- 1.52mm x 2.2mm WLCSP15 packaging
- 22nm ULL Process
- 1 MB of on-chip SRAM
- 10 MB effective memory with sparsity
- Unused SRAM can be repurposed
- SPI for interfacing with host processor
- Power gating with sleep mode
- Sub-mW inference for speech, audio, and other 1-D data
Dual Sparsity
10 X 10 =
Our hardware can achieve multiplicative benefits in speed and efficiency when both forms of sparsity are present.
Sparse Weights
-
Supports sparsely connected models
-
Only stores and computes on weights that matter
-
10x improvement in speed, efficiency, and memory
Sparse Activations
-
Supports sparse activations
-
Skips computation when a neuron outputs zero
-
10x increase in speed and efficiency
Core Design
4 Core Configuration
- 512 kB on-chip SRAM per core
- 5 MB effective SRAM with weight sparsity
- 1.3 mm2 single core (22nm process)
- AXI interface
Want to learn more about SPU-001 & Our IP Design?
Specification SheetSPU Architecture
Near-Memory-Compute
- Distributes memory into small banks near processing elements to improve throughput
- Reduces data motion by performing computations close to on-chip memory banks
- Eliminates energy and memory bottlenecks caused by accessing off-chip memory
Scalable Core Design
- Core can be tiled, scaling to match needs and constraints of any deployment environment
- Targets a wide range of applications and form factors
- Digital design can be ported to other process nodes to balance performance and cost
Software
Femtosense SDK
Our SDK supports PyTorch, TensorFlow, and JAX, so developers can get started with minimal barriers. The SDK provides advanced sparse optimization tools, a model performance simulator, and the Femto compiler.
Deploying AI Algorithms to the SPU
Start with existing models : Deploy TensorFlow, PyTorch, and JAX models to the SPU without fine-tuning. Achieve high efficiency for dense models, and even higher efficiency for sparse models.