I found the presentations encouraging. One thing that that I want to emphasize that was mentioned in the presentations...is the issue of parallelism. There seems to be a common perception that multiple parallel channels of 4, 8 or even 16 channels will solve the problem of obtaining ever higher data rates. That may be true to a limited degree, BUT there are big issues in getting highly parallel systems to work. A few things to keep in mind:
1. Each time the power through a particular channel is halved, you get 3dB optical loss, which translates to 6dB of RF loss (RF loss is always 2X optical loss)
2. Two channels have half the power of one channel, four has half the power of two, eight is half the power of four, etc.
3. Consequently, multiple channels have much lower power (dynamic range) than few channels.
4. This is compounded by higher order modulations like PAM4 and PAM8, etc. The spacing between logic levels is lower for higher order modulations, and therefore, is more susceptible to perturbations in signal levels, resulting in significantly higher bit-error-rates (BER).
For these reasons, I believe a practical limitation to parallelism will be 4, possibly 8 channels. The losses and error rates of large number of channels will simply be too high to overcome. This is the big limitation to SiP IMO. SiP only solutions will find it very hard to scale the performance required to hit the higher datarates.
Market Data 
Markets 
AI
