I think this book will give the current technical trends in hardware and software co design in embedded systems which was published in 2021.

"Embedded Systems - a Hardware-Software Co-Design Approach"

It is definitely relevant today because of Deep Learning. Most deployable implementations in embedded systems are completely rigid and depend on specific libraries provided by manufacturers such as Qualcomm, Samsung etc

Some people are betting that FPGAs will solve it all, but that may take a while and most software today can't easily change configurations on the fly, you usually have to train a different model for each target hardware.

The HW/SW partitioning for heterogeneous computing is a very relevant topic nowadays, specially in Embedded Systems. And you can take advantage of the HW/SW partitioning for problems where you need to accelerate the computation, specially if traditional sequential computation, done with microprocessors, is not enough. You would need to understand the problem application, identify which part of the problem can be accelerated, and which part, due to its nature cannot be accelerated (basically because of timing dependencies). The accelerated part of the problem may be solved by using an FPGA, and you may combine the traditional sequential computation using a microprocessor. Fortunately, there are FPGAs that have hard core processors embedded in the FPGA fabric, where you may have the case of dual or four core ARM-type processors, and have communication with the programmable part of the FPGA to accelerate part of the process. Example of entry level FPGAs of this type are Cyclone V SoC from Intel or Zynq 7000 SoC from Xilinx, both with dual core ARM Cortex-A9.