Networking

Table Memory + Packet BufferWe all love the internet, but you would never know it by looking at the memory market or, for that matter, the semiconductor market as a whole. For the last 40 years and to this day, IC development has been heavily biased towards the compute market. Whether it was the mainframes of the ’60s or the minicomputers of the ’70s, the PCs of the ’80s and ’90s or the smart phones, portables, and servers overtaking the world today, memory and processor architectures continue to mainly focus on computing. The reason is simple; unit volume. Certainly, there have been lots of IC products developed to serve the networking market, but none of them can produce the volume of sales seen in computing. The reason is simple. We may all carry around a phone but networking equipment is necessarily a shared resource. As the up-front costs of developing new ICs continue to rise at unprecedented rates, the networking market is faced with a terrible problem. The cost of advancing networking semiconductor technology may outstrip what the market is able to pay. Di3D technology offers a viable answer to that problem.

Although the ability of Di3D technology to help get development cost budgets under control is great, the real payoff for the networking market is performance. Tezzaron’s Di3D technology directly enables radical new performance improvements at an architectural level. While the compute market has struggled for decades to make parallelism work well, parallelism is the natural approach to networking. But parallelism is an interconnect-intensive proposition and that is where Di3D really shines. Memory developed with Tezzaron’s Di3D technology can deliver what networking systems need most: high transaction rates – made possible by the high granularity architecture that becomes possible with dis-integrated design techniques. And the rest of the header processing hardware needed to make networking equipment really sing can benefit from the same technology. The performance gains available are directly proportional to how bold networking chip architects are willing to be. Somebody is going to win big with Di3D.