SWCU195A December 2024 – May 2025 CC2744R7-Q1 , CC2745P10-Q1 , CC2745R10-Q1 , CC2745R7-Q1 , CC2755R10
The device includes a generic mathematical hardware accelerator, the Algorithm Processing Unit (APU), provided to offload numerically intensive operations from the system CPU. The APU operates on floating-point numbers using IEEE 754 single precision format, and it is optimized to work with complex numbers. APU operation is autonomous from the system CPU. A local memory for data storage is provided for the APU, and this memory is also accessed by the system CPU. The APU is capable of handling vector and matrix operations efficiently and can sustain a complete multiply-and-accumulate operation on complex numbers in each clock cycle. Usage of the APU as an accelerator for the system CPU significantly reduces the runtime for computational-intensive algorithms such as MUSIC (Multiple Signal Classification) needed to support the Channel Sounding feature of the Bluetooth standard, and also signal pre-processing algorithms to support Edge Machine Learning use cases. This allows the system CPU to offload applications to the APU, to then perform other high-priority tasks.
The main features of the APU are the following :
8 KB zero-latency local data memory with a data access width allowing:
Memory read bandwidth of 12.288 Gbps
Set of pipelined hardware accelerators:
Real-to-complex number converter
Fixed point to floating point converter
Floating point multipliers
Floating point CORDIC function
Floating point adder and accumulator
Max-min function
Floating point divider
Sine and cosine operations
Peak FLOPS performance of 0.672 GFLOPS