From link:

"On performance, chiplets encourage a heterogeneous computing approach, which is beneficial to AI. Different chiplets can utilize a variety of computing components like the CPU, GPU and NPU for different purposes and AI-based workloads, with these all integrated into a single package for chiplets."

https://partners.wsj.com/nasdaq-arm/information-technology/why-chiplets-are-the-next-big-innovation-in-silicon/

------------

The NPU is a Network Processing Unit . Not mentioned here is the DPU and TPU. This is why the term "xPU" covers all the possibilities. 

Although, light source chiplets do not shrink down to nanometer scale, the best form factor will be favored when it comes to packaging light sources into 2.5D and 3D packages. Think Lego with different size parts that need to be put together into a "smaller the better" package.   

POET is on its way to "POET Inside" like Intel in the early days when everything was "Intel Inside" 

Intel (Integrated Electronics). POET's platform integrates both electronic and photonic devices.