Micron Ships World's First LPDDR5 DRAM for High-Performance Smartphones

[btarunr]

Micron Technology, Inc., today announced it has shipped the world's first low-power DDR5 DRAM in mass production to be used in the soon-to-be-released Xiaomi Mi 10 smartphone. As Xiaomi's memory technology partner, Micron provides LPDDR5 DRAM with superior power efficiency and faster data access speeds to meet growing consumer demand for artificial intelligence (AI) and 5G functionality in smartphones.

"Micron's leadership in delivering the industry's first low-power DDR5 DRAM for use in a smartphone will accelerate enablement of 5G and AI applications," said Dr. Raj Talluri, senior vice president and general manager of the Mobile Business Unit at Micron. "Our customers and partners require next-generation memory solutions, based on the latest process technology, that drive unmatched power and performance to support 5G and AI systems. Micron's LPDDR5 DRAM addresses those requirements with a 50% increase in data access speeds and more than 20% power efficiency compared to previous generations."

"We value Micron's long-standing leadership and innovation in memory," said Chang Cheng, vice president at Xiaomi Group. "Micron's LPDDR5 DRAM market-leading features ensure our Xiaomi Mi 10 smartphone will remain power-efficient while still offering incredible performance and greater stability. We believe LPDDR5 will be the standard configuration for all flagship devices in 2020."

Micron designed LPDDR5 DRAM to address the growing demand for higher memory performance and lower energy consumption across a wide array of markets, including automotive, client PCs and networking systems built for 5G and AI applications. LPDDR5 can provide more than a 20% reduction in power use compared to LPDDR4x memory.

The emergence of AI in more applications is driving the need for advanced memory solutions that enable faster and more efficient access to data. Micron LPDDR5 delivers the speed and capacity needed to feed AI engines built directly into mobile processors. These processors rely on high data rates from Micron's internal LPDDR5 memory to power their machine learning capabilities.

Micron's next-generation LPDDR5 memory is designed to meet the demands of 5G networks, which will start deploying globally at scale in 2020. Micron LPDDR5 allows 5G smartphones to process data at peak speeds of up to 6.4 Gbps, which is critical for preventing 5G data bottlenecks. This capability addresses other emerging technology needs, such as those in autonomous vehicles that require a memory subsystem with a larger bandwidth to support real-time computing and data processing.

Micron is shipping LPDDR5 to customers in capacities of 6 GB, 8 GB and 12 GB and at data speeds of 5.5 Gbps and 6.4 Gbps. In the first half of calendar 2020, Micron LPDDR5 will also be available in a UFS-based multi-chip package (uMCP5) for use in mid- and high-tier smartphones. The LPDDR5 in a multi-chip package will also provide longer battery life and higher bandwidth to enable high-performance image processing that was previously reserved for flagship smartphones.

To learn more about Micron's low-power DRAM portfolio, including LPDDR5, visit this page.

View at TechPowerUp Main Site

 

[Valantar]

Hm. Would be very nice if next-gen mobile CPUs/APUs got around to including LPDDR5 controllers, though given that we're only just now seeing LPDDR4x in that space I guess that's optimistic. Still, that would make for some nice iGPU performance improvements.

 

[Steevo]

Hm. Would be very nice if next-gen mobile CPUs/APUs got around to including LPDDR5 controllers, though given that we're only just now seeing LPDDR4x in that space I guess that's optimistic. Still, that would make for some nice iGPU performance improvements.

IGPU performance would be far better with a stack of HBM next to it.

 

[Chomiq]

IGPU performance would be far better with a stack of HBM next to it.

Yeah but what's the cost comparison?

 

[R0H1T]

Yeah but what's the cost comparison?

Depends on the number of stacks, as well as the speed of HBM you'd like to pair it with. Did I mention the memory controller? Wild guess but you'd probably get the cheapest HBM at around 2x the cost of the cheapest LPDDR5 solution, a lot of the issues with HBM have to do with the fact that not only is it expensive to make but also package with something like a GPU.

 

[IceShroom]

Hm. Would be very nice if next-gen mobile CPUs/APUs got around to including LPDDR5 controllers, though given that we're only just now seeing LPDDR4x in that space I guess that's optimistic. Still, that would make for some nice iGPU performance improvements.

It would be great for APU, but who will buy AMD APU when we will have Intel iGPU.

 

[Aquinus]

Depends on the number of stacks, as well as the speed of HBM you'd like to pair it with. Did I mention the memory controller?
Wild guess but you'd probably get the cheapest HBM at around 2x the cost of the cheapest LPDDR5 solution, a lot of the issues with HBM have to do with the fact that not only is it expensive to make but also package!

On the other hand, it's very space efficient for the capacity and speed which is advantageous for mobile devices.

 

[Logoffon]

>Mi 10
What's next? Mi 11 with 3D XPoint storage connected via NVMe from Micron?

 

[Valantar]

IGPU performance would be far better with a stack of HBM next to it.

Obviously it would, but then again HBM would be massive overkill, and very expensive - especially for an iGPU-class product. I'd guess it would carry at least a $100 premium over any LPDDR5 solution, and for a relatively minor performance gain at that. An iGPU doesn't need >450GBps after all. Also, does Windows support HBM for system memory at all? If not, you'd be talking a much larger premium as you'd be moving from fast shared memory to two pools of separate fast memory. Not to mention that any and all space savings would disappear.

While I would love a thin-and-light with 1-2 stacks of HBM2E, 16-32GB capacity, I sincerely doubt such a system will ever exist.

It would be great for APU, but who will buy AMD APU when we will have Intel iGPU.

...anyone wanting superior GPU performance? Sure, Ice Lake is competitive in some applications (Intel's drivers are extremely variable, only properly optimized for a selection of popular titles) with currently available 3000-series APUs, but will be surpassed by a significant margin by 4000-series APUs in the next couple of months. Then Intel will respond with Tiger Lake Gen12/Xe iGPUs, and AMD will respond again with Zen3+RDNA APUs. My money is definitely on the latter being both faster and more efficient. On the CPU side it might be close (Ignoring node differences, Zen2 is ~6% IPC ahead of SKL and its derivatives, making ICL about 10% faster than that again (~18% over SKL), with Zen3 rumored for a ~15% improvement over Zen2 again putting them slightly ahead, and TGL rumored to surpass ICL but being too far into the future to have plausible rumors even; this is a difficult call), but on the GPU side AMD has a massive advantage of experience and driver development over Intel. Xe might be good, but it might also be entirely mediocre. And Intel has yet to get their 10nm process under proper control (just look at availability of ICL laptops vs. CML), while AMD will be moving to TSMC 7+ (in volume production currently) and 5nm (ramping up; Apple is moving to it this year meaning volume production around summer at the latest) in the next couple of years. Intel isn't quite falling behind, but they have to get their act together to keep up.