The matrix multiply accelerator (MMA) provides the following key features:
- Support for fully connected layer using matrix multiply with arbitrary dimension
- Support for convolution layer using 2D convolution with matrix multiply with read panel
- Support for ReLU non-linearity layer OTF
- Support for high utilization (>85%) for typical convolutional neural network (CNN), such as AlexNet, ResNet, and others
- Ability to support any CNN network topologies limited only by memory size and bandwidth
- Coupled with C71x CPU using DSP chassis for data formatting
- MMA and C71x CPU are on a shared power domain
- MMA cannot be independently powered off or clock gated at LPSC level (although it has extensive clock gating within the IP)