Xilinx Model Composer also supports a half
precision floating point data type, which is 16 bits wide instead of 32 bits, because it
consumes less real estate on the target device when implemented on the FPGA. This is an
important consideration when designing in Model Composer. However, Simulink does not support the half
data type, and this can lead to
errors in simulation. The solution is to use the Model Composer DTC block to convert the
half
into the single
data
type supported by Simulink, for those portions of the design that are not
in the Model Composer sub-module and will not be part of the generated output.