Advanced Micro Devices is currently the only company that uses multilayer memory with high throughput for commercial products. Although the use of HBM is an indisputable competitive advantage of AMD over rivals, the company in every possible way encourages the introduction of a new type of memory with rivals. In addition, AMD does not plan to charge licensing deductions for its intellectual property.
HBM memory has been developing at Advanced Micro Devices, SK Hynix and their partners for more than eight and a half years old. During this time, AMD has developed a huge number of technologies and principles that formed the basis of the new memory. In particular, the developers created the concept of multi-layer memory with high throughput, inter-block compounds for memory layers (Through Silicon Via, Micro-Bump, and T.D.), a layer of basic logic for the HBM (Base Logic Die), the connecting substrate (Silicon Interposer), the ultra -stroke memory of the memory in processors and much more.
Now that HBM is ready and is commercially made, AMD needs to reduce the cost of such memory that it is possible exclusively with large -scale production, which will begin only when HBM is used by a large number of products. The whole industry will benefit from reducing the cost of memory with high throughput.
NVIDIA Corporation has repeatedly announced the intention to support the HBM type, and earlier this year said that the Pascal graphic processors in 2016 will be able to use multi -layer memory with high throughput. It is believed that the second -generation HBM will be used in conjunction with the Pascal generation graphic chips to build graphic adapters and supercomputer accelerators from 8–32 GB of memory on board. HBM promises to become one of the key innovations Pascal, which will triple the memory capacity available for GPU compared to the solutions of the current generation.
Earlier this week, one of the Chinese websites said that NVIDIA will delay the use of HBM for a year due to AMD requirements to pay for the use of its intellectual property associated with a new type of memory. In particular, 2.The 5D-final of the graphic processor Fiji uses a number of AMD patents, as a result of which NVIDIA will need to either develop an analogue that do not use the competitor inventions or pay AMD deductions. The Advanced Micro Devices itself categorically denies the requirements.
«AMD does not participate in any fees of licensing deductions for HBM», — said Ian Bristow, representative of AMD. «We are actively encouraging the widespread use of HBM and related technologies [used] on [Radeon R9] Fury and do not charge intellectual property fees.
Advanced Micro Devices owns a number of patents covering HBM. Since this intellectual property is part of the Jesd235 standard Jedec consortium, the company is obliged to provide the possibility of its use for free, or on «reasonable conditions free from any unfair discrimination». In addition, it should be borne in mind that a wide cross-licensed agreement has been signed between AMD and NVIDIA, which largely prevents the collection of deductions.
A possible delay using the second -generation HBM NVIDIA can be associated not so much with the intellectual property of a competitor or a third party, but with the availability of HBM2 microcircuits. If Samsung Electronics and SK Hynix start the production of new chips only in the second half of the next year, this means that the new memory will become available only in 2017.