In 2003, we organized and presented one of the earliest conferences on new and emerging memory technologies. At the conclusion of those presentations, one of the first questions from the audience seems to have been repeated in every conference on this topic since then. “When will the cost per bit of the new technology compete with that of DRAM?” There appears to be something mystical about that DRAM cost per bit crossover point.
I also recall NOR’s introduction in 1988/89, which was accompanied by charts predicting that magic moment when NOR would dip below the cost/bit of DRAM and dominate the memory market. Even though that particular crossover didn’t happen, NOR clearly found a niche and is still here.
On the other hand, NAND prices have gone below DRAM prices as shown below. NAND SSDs are widely accepted in Enterprise servers, and we can read articles touting that “NAND is now replacing DRAM in servers.”
Cost/GByte of Memory Technologies
Source: Convergent Semiconductors 2012
However, I believe that asking the manufacturers of new and emerging memory technologies about the cost crossover with DRAMs is the wrong question—for two reasons. The first reason is because that comparison is only mildly relevant, because OEMs consider both the price AND the performance. For example, NOR still supports some applications even though the anticipated/bit crossover with DRAM was never achieved.
The second reason the cost per bit comparison with DRAM is not particularly significant is that one technology doesn’t necessarily “replace” an older one as much as it enables a completely different solution. In the case of SSDs, NAND isn’t actually replacing DRAM so much as it is enabling a new level of cost/performance in the architecture. OEMs seized the lower power consumption advantages of non-volatile memory relative to DRAM along with the faster access time relative to HDD, and from that perspective the system-level engineers modified the server architecture to take advantage of those cost/performance opportunities.
The important question to ask the memory manufacturers is: what is going to happen as memory processes proceed to less than 20nm line widths and as memory companies contemplate the conversion to 450mm fabs?
The opinion of many observers is that striving to perpetuate “Moore’s Law” is becoming progressively more expensive for manufacturers of charge-storage memory cells. The relative value of a new technology begins to improve as older technologies face increasing challenges in maintaining the historic decline in cost/bit. At the same time, the ever-decreasing number of memory manufacturers is a stark reminder of the increasing difficulties of memory companies hoping to reap the benefits of high volume production. There are probably enough highly talented engineers so that the industry can avoid falling off of the looming technology cliff, but the path ahead is clearly an increasingly difficult one marked by unprecedented levels of investment along with potentially more consolidation.
How does one get an accurate picture of where the industry is in terms of bringing new memory technologies into production?
We believe that the spreadsheet skirmishes among the new and emerging memory technologies are over, and a more mature phase is becoming evident. OEM evaluations of several of the new and emerging memory technologies are already taking place, such as Samsung’s brief experiment in late 2011 with phase change memory in cell phones. One of the older and more established companies, Ramtron, has agreed to consider an unsolicited bid from Cypress that was recently received. Crocus, a French-based memory company with primary operations in Santa Clara, is receiving financial support from RUSNANO, a Russian investment fund. Adesto’s most recent round of funding in late 2011 was lead by Altis Semiconductor, a privately held company that was previously an advanced research facility for IBM and Infineon based in France.
Instead of asking the memory companies when the cost per bit of their technologies will compete with DRAMs in some 20-year old application, we think it would be more revealing to ask OEMs what sort of new system-level improvements are potentially on the horizon that might be enabled by new and emerging memory technologies.
Bob Merritt is a founding partner of Convergent Semiconductors, LLC. He is a 25 year veteran of semiconductor and communications markets. Starting in the semiconductor industry as a sales engineer for Intel, Mr. Merritt subsequently held senior executive positions in marketing and operations at major semiconductor memory companies including Mostek, Hitachi, Siemens, and Hyundai.