Micron Technology tomorrow is set to disclose a hybrid memory technology that it claims will boost performance 20-fold over the memory chips used in PCs today.
Micron, the largest manufacturer of memory chips in the U.S., says the "Hybrid Memory Cube" can tap into the full performance potential of DRAM--or dynamic random access memory--resolving a longstanding problem referred to as the "memory wall."
Targeted initially at networking and high-performance computers, the technology will be rolled out at an investor conference in Phoenix, Ariz.
"Where DRAM is positioned in the system, you really get into some intractable performance bottlenecks," said Brian M. Shirley, vice president of DRAM Solutions at Micron, in a phone interview on Tuesday. "As you move to new memory technologies, the computer is not able to take advantage of that extra performance. This is true for not just Micron but for everyone (all memory chip makers). It's something called the 'memory wall,'" he said.
Essentially, the performance of DRAM is constrained by the capacity of the data channel that sits between the memory and the processor. No matter how much faster the DRAM chip itself gets, the channel often becomes a choke point for data.
Today, DRAM technology used in PCs is most commonly referred to as DDR3, or Double Data Rate 3.
"There's a growing wall between the kind of performance we can get off the DRAM and then getting all of that data over to the processor itself. If you look at the move recently from DDR2 to DDR3 memory, it was the first transition that I can remember where the desktop and notebook guys actually delayed their transition because in terms of real, absolute performance, they weren't seeing a big enough benefit from DDR2 to DDR3," he said.
Shirley continued. "We've rearchitected in a way to get through this memory wall and deliver a staggering amount of DRAM bandwidth directly to the processor," he said. "20X. Those are real numbers. A credible, defendable number. And there's room to grow on top of that."
The secret sauce is the memory controller (see graphic) that's been added to the memory. "By putting this logic layer--which is actually a controller chip--we were able to overcome that bottleneck by crafting a higher speed bus that will go from the [controller] chip to the CPU. A very, very high-speed bus," he said.
Micron is currently working with high-performance computing and networking companies but, like most high-performance technologies, this is expected to work its way to the consumer space in some form. "We would see this working its way to commercial (corporate) solutions as early as 2012, with significant volumes in 2013. These kind of technologies will start to work their way toward the consumer space in 2015, 2016," he said.
For now, the performance needs are most dire in networking and cloud computing. One-hundred gigabit Ethernet routers and switches and cloud computing servers require "everything they can get," he said. "This is our way of giving them a fire hydrant."
Micron, unlike the multitude of "fabless" semiconductor design houses in the world, is also one of the world's largest manufacturers of memory chips, with large plants in Idaho, Utah, and Virginia, among other locations. Like Intel--with which Micron has a flash-memory chip manufacturing joint venture--Micron actually builds the technologies it develops.
Customers will include major processor suppliers. Though no customers have been named yet, it would probably be safe to assume that one of those companies is Intel, the largest chipmaker in the world.
Micron will also continue to work on next-generation memory technologies for the consumer space, such as DDR4, Shirley said.