TAIPEI Taiwan -- In an unlikely turn that wouldn't be out of place in the funny papers, a comic book company will field a microprocessor in China that pointedly limits the use of intellectual property from the West.
With the speed of a Bruce Lee karate chop, Hong Kong-based Culturecom Holdings Ltd. is shedding its past as a hawker of popular kung fu titles, like Dragon and Tiger Heroes, and springing forth to enter the dragon's market (see July 31 story) .
Culturecom, which enjoys close ties to the Chinese government, has launched an IBM PowerPC-based system-on-chip equipped with a unique Chinese character-generating algorithm. Quicker than you can say "hai yah!" the SoC seems poised to pull ahead of competing systems by dint of using only 256 kbytes of on-chip memory to interpret more than 32,000 characters.
The ViaDragon, or V-Dragon, and its embedded code run in a Linux environment " a combination that is being keenly watched by Chinese officials who want to avoid reliance on intellectual property from Western companies, such as Microsoft and Intel, as China plows big bucks into modernizing government, education and state-owned industrial institutions.
"The government is using our chip to fight Microsoft. They think that Microsoft is too much of a monopoly, and they cannot allow a monopoly situation by Western manufacturers in China," said Benjamin Lau, senior vice president of Culturecom.
Culturecom claims its algorithm allows the chip to generate at least 32,000 Chinese characters, in both the traditional and simplified forms used in Taiwan and China, respectively. Its creators say the ViaDragon architecture can be extended to more than 50,000 characters, in three fonts and a range of sizes, and can interpret other Asian languages, such as Korean and Japanese.
The company is hoping that the mix of extensibility and meager memory will attract mobile-system makers, a massive market in China, along with network PC makers selling intohigh-volume government jobs.
Out of comics
In its current form, Culturecom has existed for only about four years. Before that, it was a widely known publisher of kung fu comics. The company still derives revenue from that business but has now drop-kicked into a tech firm, thanks to a group of overseas Chinese investors who bought out the comics business, relegated it to a side venture and kept the name because they liked it. Lau said buying an existing company and gutting it was easier than setting up a new business in Hong Kong.
Though young, Culturecom has already shown its determination to be a serious player. It has parlayed a close relationship with China into an exclusive deal to sell electronic-book readers, priced at roughly $72 and subsidized by the government, to all of China's middle- and high-school students " about 200 million kids. The readers are considered a test system for the first iteration of ViaDragon, its embedded algorithm and the Linux Midori operating system that Culturecom recently licensed from Transmeta Corp.
Culturecom is hardly the first to come up with a Chinese-character-generating algorithm. Its version uses the popular Can Jie coding system, a way of separating characters into their base components and assigning numbers to them that was developed by Chu Bong Foo, a 20-year veteran of Chinese character-coding systems and the vice chairman of Culturecom.
There are basically three ways to generate Chinese characters. The ViaDragon employs a stroke-based approach (Chinese characters were originally drawn with brushes), rather than an outline or character-encoding system. A stroke-based algorithm saves memory space, important for manufacturers of small-form-factor and mobile devices.
In contrast, a character-encoding method uses a lookup table of bit maps. The method is fast, but a 24 x 24 font for roughly 7,000 simplified Chinese characters needs 500 kbytes. Also, Chinese character bit maps don't scale that well.
The outline method, meanwhile, defines a set of points along the outside of the character and fills in the space to create it. An outline font can more easily scale to different sizes than character encoding, but the trade-off is using up 2.5 to 3 Mbytes for 7,000 simplified Chinese characters.
Compression and look-up tables
The Culturecom architecture's extensibility seems to hinge on a combination of compression techniques and cleverly devised lookup tables, which reduce duplication for different fonts and strokes by accessing a set of common rules, according to David A. Smallberg, a senior lecturer in computer science at the University of California-Los Angeles. With a small team of colleagues, Smallberg, a specialist in software internationalization and Asian languages, analyzed the ViaDragon last year.
"The combination of efficient encoding, compact internal representation, 'smart fill' for strokes and other techniques will be difficult to duplicate," wrote Smallberg in a report on the findings. "The ability to generate tens of thousands of characters in minimal space, efficiently and with pleasing aesthetics cannot currently be accomplished by any other technique."
Besides pushing ViaDragon, which will be offered as the IA3210, to government and industry, Culturecom is eyeing commercial markets such as smart toys or other platforms that use an LCD. At the chip's official coming-out party in Hong Kong about two weeks ago, the company said it had orders for 600,000 units, which would go into network PCs in factories and government offices.
Although Culturecom is pushing a platform, including a reference design for network PCs, it acknowledges that the greatest asset is not hardware but the embedded algorithm. Culturecom outsourced SoC development to IBM Corp., which will manufacture the chip in an 0.18-micron, five-metal-layer CMOS process. Lau said the company is willing to license the algorithm and is in talks with several vendors to do so. One, he indicated, is BLX IC Design Corp., the government spin-off that designed China's first CPU, Godson-1.
In January, the second-generation V-Dragon, better suited to run multimedia apps in a PC environment, will roll out, he said. Japanese and Korean versions of the algorithm will be ready by the end of this year.