Fit multimedia handsets with the right memory architecture
Arie Tal, M-Systems
Jun 09, 2005 (5:00 AM)
3G is here, and it's one of the most talked-about, high-profile topics in the mobile and wireless industry. 3G offers potential benefits to users, operators, and handset makers and, according to one industry analyst, 42 3G networks were operational as of June 2004. With every passing day, announcements in countries around the globe are being made, introducing operational 3G networks. By 2007, about 25% of all handsets shipped will be 3G-ready.
Providing a bandwidth of 384 kbits/s in current UMTS networks and over 8 Mbits/s if upgraded to HSPDA (also known as 3.5G, and already defined in the 3G standard), the 3G network is approaching the speed of ADSL connections. This wide bandwidth unplugs a major bottleneck, specifically regarding the network's inabilities to handle the huge amounts of data required for multimedia content. Until now, this bottleneck has prevented many applications and services from materializing.
Handset makers are also at a crossroads. The camera phone was a catalyst for the dramatic growth in handset sales in 2004 and 2005. So much so that in Japan, camera phone penetration is close to 100%. While the global average in 2004 lags far behind, with only some 30% penetration, analysts are predicting this number to rise to 60% by 2007. However, studies show that this year, the growth rate will slow as a new killer feature is sought to replace cameras in phones to drive the handset replacement market.
To overcome this slowdown, handset makers are adding more multimedia capabilities, turning them into portable entertainment stations. In fact, some handsets already offer a full range of multimedia services. They can play MP3s, record video and screen-selected TV programs, run 3D games, and take high-resolution still pictures.
With 3G up and running, handset makers are releasing multimedia devices that will drive up replacement sales, while at the same time enable operators to provide band-consuming content to boost operator revenues. A clear indicator of this win-win trend can be seen with the formation of alliances between content owners and handset makers and operators. Recent examples include the Motorola-MTV and Motorola-iTunes agreements, as well as the Sony Ericsson-Turner Broadcasting, Samsung-Virgin Records deals.
Driving changes in the architecture
To enable this trend, handset architectures are undergoing a major redefinition that's only partially noticeable to the untrained consumer eye. The latest smartphones, for example, are running at up to 600 MHz clock frequencies, comparable to the processing power of laptops some 3 years ago. Simple camera phones are advancing from low-end 40-MHz ARM7 devices to ARM9s running at over 100 MHz.
A third alternative uses the ARM7 to manage communications and run the phone OS, while implementing a dedicated, powerful multimedia processor to handle the influx of multimedia content. This approach is favored by many handset vendors as a transition measure, enabling them to use the same, main processor and thereby benefit from reduced costs, shorter development cycles and more efficient inventory management.
More memory
One of the most dramatic effects of the multimedia revolution on handsets is in the amount of memory they carry, especially non-volatile memory (NVM) for running and storing code and storing data. Just three years ago, the average handset carried 4 Mbytes of storage. Today, capacities are at 64 Mbytes and higher. In fact, Samsung's recently announced SPH-V5400 contains a 1.5-Gbyte hard drive.
Higher capacities are only part of the memory-for-multimedia story. Multimedia applications running on small, battery-powered handsets challenge memory performance, power consumption, and size. Despite these challenges, and often at odds with them, memory reliability remains a critical requirement. The same handset that runs multimedia must be always on and/or ready in an emergency to serve as a simple phone/phone book.
While waiting for 3G to come of age to provide the bandwidth to meet multimedia requirements, memory makers have introduced multiple flash and mechanical hard-drive technologies, marketing them in a wide range of embedded and removable memory products. Handset designers in need of increased NVM and high performance must sort through all these offerings.
More flash choices
NOR- and NAND-based flash memory are vying for market share, each backed by huge companies such as Intel and Samsung. Recently, hard-drive vendors have begun targeting the music handset market in an effort to completely displace flash technology. NOR technology, the older and more entrenched of the two, is better known to most engineers and more reliable than NAND technology. NAND-based memory uses less silicon and is therefore more cost-effective. Note that the first NAND to be implemented in a handset was a NAND-based embedded flash drive (EFD). With a built-in controller that uses a standard NOR interface, the EFD offers XIP boot capabilities so that it can run code as well as store data, implement error detection and correction (EDC/ECC), and provide protection and security-enabling features for digital rights management (DRM).
EFDs were critical to the successful penetration of NAND into the handset market. Today, they co-exist with raw NAND die. Together, they're consistently driving NOR into the lower ends of the handset market where less memory is required, while becoming the default memory for multimedia handsets in general and in smartphones in particular.
To better understand where and how NAND, NOR, or EFD is used in handsets, it's important to understand the dynamics of the handset segments and its ecosystem and architectures. A number of factors have joined forces to create the right ecosystem in smartphones to support NAND flash and EFDs.
Smartphones target business users, in need of devices to increase their out-of-the-office productivity. These devices usually combine a handset and a PDA, with a range of additional functionality. Smartphones feature large screens, a powerful OS, such as Windows Mobile, Symbian, PalmOS, or Linux, e-mails applications, and Office-like software. In addition, smartphones include games, a high-resolution camera (for still images and video), and enough memory to store and play MP3 songs.
Today, a typical smartphone packs anywhere from 32 to 256 Mbytes (and growing) of embedded memory. In this capacity range, NOR is too costly to compete with NAND. However, the first smartphones that incorporated NAND or EFDs used it as a disk-like solution to supplement rather than replace the embedded NOR flash. This approach was necessary because raw NAND can't execute-in-place (XIP). Although NAND-based EFDs do offer XIP boot, they require additional SDRAM as the OS is copied to it and runs from it (similar to a PC architecture). This reduces the cost advantage over a NOR-based architecture. To circumvent this need for more SDRAM, applications were stored and executed on the resident NOR flash, and user data was kept on the NAND media.
Smartphone OS vendors have met this challenge by adding paging capabilities to their OSs. Thus, instead of shadowing the entire OS image to RAM, only the kernel is copied, while applications are paged in and out of the RAM when needed. This is known as "paging on demand" or "store and download." Paging on demand has dramatically reduced the amount of SDRAM needed in NAND-based devices to about half (from 64 to 32 Mbytes) and maximized its cost benefits. Today, most smartphone designs implement some kind of NAND technology as an embedded NVM solution, replacing the former NOR-based architecture.
Feature phone needs
Feature phones constitute the majority of the mobile handset market. They're standard phones enhanced with features to provide entertainment. The killer application of feature phones in 2004 was the camera, which enable users to not only take pictures and video clips, but ostensibly to send them as MMS messages to other users (incompatible network standards don't fully support this extended capability).
To support this enhanced feature set, feature phones have undergone architectural changes. Today, two architectures dominate this category, enhanced and multimedia-centric (see the figure). The enhanced legacy architecture offers only a slight improvement over basic voice-centric designs. It introduces additional flash media for storage and strengthens the baseband from an ARM7 to an ARM9. This architecture can support basic camera functions and simple Java games, but can't support advanced video capabilities and 3D gaming.
Two common feature-phone architectures are the enhanced legacy (top) and the multimedia-centric (bottom).
A dedicated multimedia coprocessor was introduced in the multimedia-centric architecture. The baseband continues to use a relatively weak ARM processor (usually ARM7), and is responsible for all traditional handset functions and the OS execution. The multimedia coprocessor, in contrast, is based on an ARM9 and higher, and includes additional multimedia-targeted state machines. This processor handles all multimedia and graphic-intensive applications, and is used on demand of the baseband. This architecture typically includes two memory subsystems, one for the baseband, to store and execute the OS, communications stack and applications; and one for the multimedia coprocessor, to store its code and all files managed by the file system (mostly multimedia). Although different, these two architectures both require flash media to store pictures, video clips, and games downloaded by the user, but capacity requirements differ depending on the phone segment.
Low- to mid-range feature phones
Low- to mid-range feature phones provide a basic, low-res camera (VGA or lower) and gaming. A low-res still picture only occupies 40 kbytes or less, a requirement that NOR can meet. Many feature phones provide 32 Mbytes of NOR, the upper limit of NOR as a competitive media, and in many cases an additional slot for a NAND-based removable card (mostly SD or MMC form factor).
For these devices, NAND media offers a marginal cost improvement, which many vendors don't find compelling enough to warrant an architecture change. As photo resolutions increase, a growing demand for built-in NAND media is evident. NAND will first serve as a disk-like solution only, because paging software isn't yet available in the real-time OSs used in these handsets, resulting in increased PSRAM size and reducing the NAND/MLC NAND cost advantage. However, to further reduce the BOM while increasing the storage capacity, memory vendors are joining forces with feature-phone OS vendors to introduce OS paging capabilities. Once ready, this capability should dramatically accelerate the penetration of bootable NAND into handsets as the only NVM media on board, serving both as a code and storage media. It's important to remember that most chip sets for feature phones and most OSs don't support raw NAND, and thus designers turn to EFDs for NAND support.
Mid to high-end feature phones
At the upper end, the feature phones offer a VGA or higher-resolution camera (the mainstream camera for these devices in 2005 is 2 Mpixels), and are capable of video recording, some image processing and advanced gaming. These devices are based either on the multimedia coprocessor architecture, or on a higher performance baseband processor (ARM11) that's robust enough to manage multimedia. In such handsets, NOR media is not a viable solution, as evidenced by the fact that there's massive penetration of NAND flash technology. The most basic function of NAND in these devices is that of a disk drive, much like the hard drive in a PC. However, many handset vendors are taking advantage of EFDs to eliminate the additional NOR device to store the multimedia processor code.
About the author
Arie Tal is the director of marketing for M-Systems’ Mobile Division. He received an MBA from Haifa University and a BA in economics from The Technion, Israel Institute of Technology. He also has a practical engineering degree in Electronics from Netanya College. Tal can be reached at Arie.Tal@m-systems.com.
Related Articles
- Choosing the right memory for high performance FPGA platforms
- Time for multimedia SoCs to get their analog signals right
- Choosing the right A/D converter architecture and IP to meet the latest high speed wireless standards
- A Performance Architecture Exploration and Analysis Platform for Memory Sub-systems
- Selecting the right Nonvolatile Memory IP: Applications and Alternatives
New Articles
- Quantum Readiness Considerations for Suppliers and Manufacturers
- A Rad Hard ASIC Design Approach: Triple Modular Redundancy (TMR)
- Early Interactive Short Isolation for Faster SoC Verification
- The Ideal Crypto Coprocessor with Root of Trust to Support Customer Complete Full Chip Evaluation: PUFcc gained SESIP and PSA Certified™ Level 3 RoT Component Certification
- Advanced Packaging and Chiplets Can Be for Everyone
Most Popular
- System Verilog Assertions Simplified
- System Verilog Macro: A Powerful Feature for Design Verification Projects
- UPF Constraint coding for SoC - A Case Study
- Dynamic Memory Allocation and Fragmentation in C and C++
- Enhancing VLSI Design Efficiency: Tackling Congestion and Shorts with Practical Approaches and PnR Tool (ICC2)
E-mail This Article | Printer-Friendly Page |