There isn't even an official standard for DDR4, the next generation of computer memory technology. But memory manufacturers are already shipping samples of the first DDR4 memory units, and preparing to produce them en masse. On May 7, Micron joined the field, announcing it had released its first fully functioning DDR4 memory product for testing.
Micron says its product, a 4-gigabit x8 DDR4 memory unit developed in partnership with its Taiwanese partner Nanya, will ship on a variety of memory module configurations by the end of the year. (Micron also announced that it is also developing x16 and x32 DDR4 memory components, but didn't give a timeframe for them.) And Micron's competitors in the space are also preparing to ship their own DDR4 modules in that time. That means that memory modules based on the faster, more power efficient technology could start shipping on servers (where its benefits are in the greatest demand) by 2013—if there are any CPUs ready to handle them.
What is DDR4?
The Joint Electron Devices Engineering Council (JEDEC) is expected to sign off on a final DDR4 standard this summer. But a draft of the spec and its key attributes were issued last August. Those preliminary specs call for DDR4 memory to operate at a maximum of 1.2 volts (20 percent less than current DDR3 memory) and achieve data transfer rates of 3.2 billion transfers per second (double that of the top-end speed of DDR3's memory bus).
Micron's initial memory DDR4 units will handle transfer speeds of 2.4 billion memory transfers per second, with later versions ramping up to the 3.2 gigatransfers per second top-end rate defined by JEDEC; right now, that translates to roughly the same number of bits per second in throughput. Samsung has advertised a throughput of 2.113 gigabits per second for its DDR4.
The "DDR" in DDR4 is an acronym for "double data rate"—all DDR memory moves data across the memory bus twice for each cycle of the bus' timing clock. That's the most significant bit of architecture that DDR4 shares with its predecessor. Instead of using multiple shared channels to link memory units with the CPU's memory controller, each DDR4 memory module has its own dedicated point-to-point connection.
Aside from the reduced voltage level, DDR4 reduces overall power requirements through an improved version of the pseudo-open drain interface (POD) logic, a technology used in graphics DDR 3. POD logic interfaces draw no current except when they're flipped to their low state—when they connect to ground.
DDR4's technical advances come at a price. The first is accomodating all of those additional memory connections. While the point-to-point architecture gives DDR4 the ability to read and write more data per cycle, it becomes a challenge to handle large amounts of memory. Servers built for DDR4 memory may use high-speed digital switches in order to reduce the number of direct memory channels connecting to the CPU's memory controller.
The other challenge is the density of the memory itself. With all those point-to-point connections, the memory chips themselves need to be very densely designed to fit in the same space. Micron's and Samsung's DDR4 memory is fabricated with 30 nanometer technology—a die density on par with the fabrication process that current CPUs use. That will make DDR4 memory, at least initially, fairly expensive to produce.
Double-teaming
Micron is a little late to the DDR4 party. Korean memory manufacturers have had the run of it for over a year: Samsung announced it had completed development of its first 4Gb DDR4 DRAM module in January of 2011, and started shipping 2Gb DDR4 samples in December of 2010; Hynix demoed its own DDR4 technology in February of 2011.
There's a lot riding on Micron's efforts to catch up and pass Samsung in DDR4 performance because of the potential gains in market share from the growing data center and cloud server market. In that area, energy efficiency and large in-memory applications are major requirements. IHS iSuppli projects that DDR4 memory will make up more than half the memory market by 2015.
Owning a big piece of that market is a gain that Micron and Nanya could use in the increasingly competitive memory space. To put itself in better position to compete with Samsung and Hynix, Micron made another announcement on May 8—it had the exclusive right to acquire bankrupt Japanese memory manufacturer Elpida Memory.
That competition could be a good thing for pushing DDR4 forward on desktops and laptops, as it pushes the manufacturers to drive down the cost of manufacturing the technology. But the gating factor for DDR adoption will be how quickly CPU manufacturers adopt the technology and integrate it into CPU memory controllers. Intel's latest statements indicate the company doesn't have DDR4 on the roadmap for its processors until the Xeon Haswell-EX processor—in 2014. Desktop and notebook users may have a bit longer to wait. And the timeframe for a low-power version of DDR4 for mobile devices hasn't even been set.
