Over the next several years, wireless carriers will rollout new multimedia services at an unprecedented pace. These media-rich services will place new requirements on the platforms that power handsets. While the semiconductor technology to enable these new services has been available for some time on large PC platforms, commercialization of this technology for wireless handsets will underscore the successful monetization of new wireless services.
Unlike the PC industry where memory plays a secondary role to the user experience, mobile memory will play a central role in exciting the senses of handset users. OEMs will be forced to re-think their platform's memory subsystem and find new memory solutions that deliver multimedia services without compromising the user experience.
Commercialization in the form of low power-high performance semiconductor products, small form factor packaging and low cost board assembly, will present a number of challenges to semiconductor device suppliers. Samsung is aggressively addressing this challenge by developing new mobile memory products which pave the road for a new breed of handsets. Two recently announced products, 512Mb OneNAND(TM) and 256Mb x32 mobile DDR DRAM, substantiate Samsung's commitment to accelerating the adoption of the new wireless lifestyle.
The transformation from wireless voice to wireless lifestyle is occurring at an accelerated pace. By 2006, analysts forecast a worldwide wireless subscriber base of 1.6B users. (Source: Friedman, Billings, and Ramsey). By 2005, Mobile Wireless Internet users will exceed PC Internet users. Moreover, mobile entertainment services will generate the largest demand by 2007.
Wireless carriers have ambitious plans for exploiting the new wireless lifestyle. However, they will need to form new business models that include strategic partnering with content providers and content creators. Unlike traditional carrier business models where voice is the dominant component of ARPU [average revenue per month], new business models will have increasingly larger infotainment components. To effect such a change, wireless carriers will need to explore creative partnering models including, but not limited to:
(1) sourcing their entertainment/news content from content owners,
(2) investigating co-branding of their wireless portals with content owners and
(3) partnering with mobile portal developers to embellish the mobile browsing user experience.
Each of these models will require sharing of revenue and capital investment with the end game being ARPU preservation and hence, sustainability of the wireless carrier industry (See Figure 1).
Figure 1 Wireless Lifestyle
A closer look at the technology required to effect the wireless lifestyle sheds light on the radical changes underway within the cell phone. By 2007, a large portion of wireless networks will have broadband [> 2Mbps] capability. This will enable carriers to offer many innovative services (See Table 1).
Table 1 Innovative Services
These new services will require a “new breed of handset” if they are to be successfully monetized. As carriers transform their networks and service offerings to usher in the wireless lifestyle, they will increasingly demand newer, flashier handset from OEMs. These handsets will ultimately become a symbolic expression of the user's lifestyle.
The current handset market can be broadly categorized into three segments: voice-data-centric, feature phone and smart phone. Current voice-centric phones have narrowband data capabilities – SMS messaging and web browsing ” and made make up roughly 40 percent of cell phone shipments in 2003. Current feature phones are camera enabled and in some instances, support 2-D gaming, Video playback, Java OS, and Bluetooth. Smart Phones feature an advanced OS [Symbian, Microsoft PocketPC, and Palm], office applications, AV streaming and remote-sync. As is the case for all high tech markets undergoing rapid change, advanced features will show up in the high-end segment first. Over time, features that are currently commonplace in the high end will migrate to the low end. Cell phones are truly at an inflection point in their evolution from voice centric to multimedia centric. Today's smart phone and camera phone will evolve into remote smart agents and entertainment agents managing our insatiable appetite for infotainment anywhere and anytime. Figure 2 shows the dramatic increase in unit growth for feature and smart phones over the next three years. This forecast is indicative of wireless carriers resolve in demanding multimedia-laden handsets (See Figure 2).
Figure 2 Smartphone predictive growth
The user experience of next generation handsets will, in large part, determine the success or failure of the new wireless lifestyle. Handset user experience will become a function of carrier's network robustness and service offering, handset ease of use, handset feature-set and handset battery life. Assuming carriers do their part in enabling the wireless lifestyle, we can turn our attention to the handset's semiconductor platform which is chiefly responsible for battery life and feature-capability. The list of next generation services identified in Table 1, sheds some light on future use trends of wireless subscribers. Figure 3 identifies various use trends and the impact they will have on memory requirements of handset.
Figure 3 Wireless subscriber use trends
Camera usage on the cell phone will extend beyond taking pictures and include photo mail, photo editing/special effects, and action shots to name a few. With 4Mp camera phones expected in 2005, images captured on the cell phone will become indistinguishable [in quality] from that of mainstream digital still cameras. To create demand for such phones, mobile processor and mobile memory suppliers will focus on optimization of photo shoot experience. Capturing, processing, compressing and storing a 4Mp image involves the shuffling of enormous amounts of data between mobile processor and memory (See Figure 4).
As image resolution grows beyond 2Mp, Mobile DRAM as opposed to Pseudo-SRAM becomes mandated, due to its lower cost/bit and higher Mbps/pin.
Figure 4 Shuffling data between processor and memory
The memory bandwidth problem is exacerbated when the feature set is extended to include multi-shot/sec capability (See Table 2). If three shots are to be taken within one second, the bandwidth requirement increases three fold. An additional data flow challenge specific to camera phones involves storing the picture. After the image is processed and compressed, it must be stored before another picture can be taken. For multi-shot camera phones, three stills must be stored before another picture can be taken. Storage time is a function of Flash program time. Figure 4 and Table 2 not only demonstrate the viability of SLC NAND technology for next generation media-rich cell phones, but they also uncover the futility of using NOR or MLC based NAND technology. The key metric being “time between shots” is a measure of Flash program time. As pixel density of image increases, so does the need for SLC NAND.
Table 2 Impact of Multi-shot image on memory
Samsung has launched a new SLC based NAND product called OneNAND(TM). OneNAND is currently available in 512Mb densities and differs from Samsung's conventional SLC NAND in its use of an integrated NAND controller and NOR interface. The integrated NAND controller enables OneNAND to achieve a sustained read performance exceeding 31MB/s. Its NOR interface allows designers to use a NOR Flash controller for transactions between processor and OneNAND. OneNAND's tremendous read performance benefits camera phone users who frequently upload their pictures to PC. Imagine uploading 128Mbytes worth of photos to your PC using your camera phones USB 2.0 port. Compared to MLC based NAND, OneNAND cuts the time by roughly 6x.
Video usage will offer many revenue-generating opportunities for carriers. Streaming video, video mail, video telephony, mobile TV, and mobile PVR are but a few. Keep in mind that as with camera usage on cell phones, users can find ways to entertain themselves without the carrier's network. A feature's ability to excite the senses without the carrier's intervention will become instrumental in accelerating adoption among users. At VGA resolutions, video-capture and video telephony stress memory bandwidth more than 4Mp image capture (See Figure 5).
Figure 5 Impact of video phone pipeline on memory bandwidth
Above 100MB/s sustained memory bandwidth, mobile DDR DRAM becomes mandated, as it provides a 2x improvement over Mobile SDR with a marginal increase in power consumption. At 100MHz clock, a mobile DDR x16 device delivers 400MB/s of peak bandwidth for jitter free video. One might ask why target 400MB/s peak bandwidth when only 100MB/s is needed for video telephony. As it turns out, the typical videophone in 2005 will have multiple processes running in parallel with video telephony. Each of these processes will steal precious memory bandwidth away from the main consumer of bandwidth ” video telephony. Add to this the challenge of designing memory controllers which can utilize 70 percent peak bus bandwidth, and it becomes abundantly clear that designers will opt for memory devices that deliver in excess of 300MB/s of bandwidth.
3-D Gaming Usage
From a business perspective, 3-D gaming may have the most attractive proposition. Revenue can be generated via sale of games [bundling, monthly download, etc] and scaleable airtime charges. Airtime charges would conceivably scale with the number of online players. More importantly, multiplayer online gaming is not a large consumer of precious network bandwidth. This enables carrier to deploy other services to the same user or other users simultaneously. For example, while carrier is collecting air time charges for multiple users engaged in online gaming, those same users might fancy a multi-way conversation and be charged additional airtime for voice. While the business proposition is very attractive, the technical challenges presented to the handset platform architect are daunting. However, the challenges are being overcome, as evidenced by the spate of recent announcements surrounding mobile 3-D hardware. [Go to the following website for more information: www.khronos.org]. Mobile chipset suppliers are developing 3-D pipelines, which will rival the quality of Playstation II. The visual quality on the current Game Boy Advance device will pale in comparison to the quality of 3-D game phones in mid 2005.
A closer look at 3-D processing algorithms reveals a voracious appetite for memory bandwidth. One, which far exceeds the bandwidth requirements of video and imaging use models. Unlike video and imaging, 3-D processing synthetically creates images in real time and the processing required to photo-realistically create scenes at 60 frames per second is mind-boggling. Furthermore, the memory bandwidth requirements exhibit tremendous variance from scene to scene. This implies that during game play certain scene sequences may require a peak bandwidth far exceeding the average sustainable bandwidth of the memory controller. As a result, 3-D architects typically specify memory devices that deliver 2x the bandwidth required for the average scene. Calculating memory bandwidth requirement for 3-D game processing has many dependencies: texture cache efficiency, depth complexity of scene, frame rate, amount of alpha blending, etc. A huge benefit is gained from improving texture cache efficiency, as texture processing [painting special effects on each pixel] is a huge memory bandwidth hog (See Figure 6).
Figure 6 Benefit of improving texture cache efficiency
For the calculations in Figure 6, a texture caching efficiency of 65 percent was assumed. Mobile DDR DRAM is clearly mandated even at paltry QVGA screen sizes. Moreover, unless mobile DDR x16 devices are stacked in a manner, which widens the memory bus to 32 bits, they too fall short of 3-D algorithms staggering bandwidth requirement.
In anticipation of this new requirement, Samsung has launched the industries first monolithic 256Mb x32 mobile DDR device [see related press release]. This product delivers 1.06GB/s of peak memory bandwidth.
While at first glance the need for fast storage may not seem evident during 3-D game play, closer inspection reveals a need for ultra-fast storage. Imagine the user, who after 30 minutes of play has a whim to pause game play for several hours, only to resume play later that day. This would require a game state save in which as much as 5MB of data must be transferred from mobile DDR to Flash. Comparing SLC based NAND/OneNAND to MLC based NAND, the case for the former is undisputed. NOR is not even considered here, as it would severely compromise user experience and battery power.
Figure 7 Impact of flash memory bandwidth on 3-D gaming
Subscribers demand for media-rich handsets will force handset designers to loosen their grip on the mobile memory status quo [SRAM and NOR] and embrace the next wave of advanced mobile memory solutions e.g. NAND, OneNAND and mobile DDR DRAM. This sea change event will open up opportunities for new entrants, as well as existing players. Given the enormity of next generation handset memory requirements, one can safely predict a lucrative mobile memory market. The stakes are high, as are the rewards. According to Web feet research, building a memory fabrication plant requires a capital investment exceeding $2B. However, the mobile memory market is pegged at ~ $12B per annum in 2007. By most accounts, this figure matches or exceeds the memory market for PC's in 2007. Mobile memory suppliers will spend large sums of money innovating their product line in order to capture a piece of this lucrative market. The end game is clear for mobile memory suppliers: dominate the emerging mobile memory market. However, the recipe for success is not as clear, as the rules have changed. In addition to large fabrication plants and access to capital, the new mobile memory frontier requires an intimacy with platform architectures and their impact on user experience.
Ivan Greenberg is the Director, Strategic Marketing, Memory Division at Samsung Semiconductor. Mr. Greenberg joined Samsung Semiconductor in 2003 and is currently the director of strategic marketing focused on the development of wireless memory semiconductors. He has more than 18 years of comprehensive experience in SoC solutions and previously held positions with GTE and BOPS. Greenberg received his BSEE from Northeastern University.