Aiming to become the global leader in chip-scale photonic solutions by deploying Optical Interposer technology to enable the seamless integration of electronics and photonics for a broad range of vertical market applications

Free
Message: A reminder of what is in store for POET

For some reason I  bookmarked this post on my phone. It was posted on July 22 last year and is agood reminder of what is in store. 

High Performance Computing for Artificial Intelligence Applications

posted on Jul 22, 2020 10:47AM

The following statement in this report really gives you a quick view of just how far reaching this platform can become. The key is to achieve both the shortest possible electrical distances which results in low impedance (reduced parasitic) in combination with the lowest optical loss with-in the same fabric (the dielectric stack). This results in the ability to achieve the highest speeds and the lowest power requirement. It is all about the dielectric stack and how it is used. The heat generated is minimized, isolated and channeled away from each die as required to remain within the required operating temperature range of each die (critically important to the optics).  Can’t be done efficiently in a stacked die configuration. It is not hard to understand why the Tier 1 wants this work kept quiet for as long as they can but at some point everyone will know.  

Optical data transmission to increase fiber efficiency is a starting point but if this platform is adopted as a standard it could expand well beyond those applications.

POET is working with a customer on is a complete system-in-package for optical computing, all built on an optical interposer. Taking advantage of a flexible platform for hybrid device integration, the optical interposer enables co-packaging of a variety of analog, digital, electronic, and photonics devices into a chip-level, self-contained computing accelerator. Its performance can match or exceed traditional processors from NVIDIA and others, but consumes a fraction of the power and can be manufactured for a fraction of the cost. By handling the mathematic computations required for Machine Learning (ML) and Deep Learning (DL) in the optical domain, rather than the digital electronic domain of a typical GPU, this technology is well suited for the low-power requirements of multiple Edge AI applications. POET’s Optical Interposer is extendable to self-contained optical systems now being developed for Artificial Intelligence applications.  

This self-contained system-in-package leverages all the advantages of the optical interposer - coupling different active devices such as lasers and photodetectors to waveguides, multiplexers, and custom siliconbased digital and analog processors. Other unique properties of the platform include the ability to remove heat from active devices and the high-speed copper and aluminum traces within the interposer, which provide the fastest available connections between all the devices within the system.

Share
New Message
Please login to post a reply