Drazen Borkovic

Breakpoints in neural network accelerator

Using vector clocks to simplify a dependency graph in a neural network accelerator

Vector clocks for highly concurrent execution engines

Static memory allocation for neural network inference

Reconfigurable neural network processing based on subgraph recognition

DMA synchronization using alternating semaphores

Efficient utilization of processing element array

Reconfigurable neural network processing based on subgraph recognition

Vector clocks for highly concurrent execution engines

Saving intermediate outputs of a neural network

Time-based memory allocation for neural network inference

Allocation and placement of resources for network computation

Configurable delay insertion in compiled instructions

Breakpoints in neural network accelerator

Efficient race-condition detection

Synchronization of DMA transfers for large number of queues

Low latency neural network model loading

Synchronization of concurrent computation engines

Neural network operation reordering for parallel execution

Efficient utilization of processing element array

Hierarchical partitioning of operators

Neural network layer-by-layer debugging

Neural network operation reordering for parallel execution

Performance debug for networks

Synchronization of computation engines with non-blocking instructions

Fast context switching for computational networks