The engineering daring that led to the first Chinese personal computer ' TechCrunch

The engineering feat that brought about the first Chinese personal computer. With 1000s of Chinese characters and limited memory the Sinotype III's inventors had to push the boundaries of early machines.China is today one of the most prosperous digital economies in the world, with an unrivalled hardware supply chain and a slew of highly profitable and prominent companies such as Tencent, Alibaba and ByteDance playing a major role in the global economy. All of this innovation is based on a 40-year old solution to one the greatest computing problems: the development and use of Chinese word processing.China began to import more computers from the West and the United States in the 1980s. In 1980, China imported 600 microcomputers made in China, compared with 130,000 in 1985. One observer said that companies in Europe, Japan, and the United States clamored to be part of this buying spree.However, this was a problem for both potential Chinese computer users as well as Western manufacturers. No Western-built personal computer or printer, monitor or operating system was capable of processing Chinese characters input or output.Both potential Chinese computer users as well as Western manufacturers had to deal with the fact that there was no Western-built personal computer or printer, monitor, operating system, or program capable of processing Chinese characters.The problem of memory, specifically the memory needed for Chinese fonts, was one of the main reasons. Western designers and engineers discovered that an English font could be created using a 5-by-7 bitmap grid, which requires only 5 bytes per symbol. This grid, although not very attractive, was able to display the Latin alphabet letters on a computer terminal or paper printout. The only 475 bytes required to store the 95 printable characters of U.S. ASCII took just 48 KB of memory, which is a small fraction of the memory needed for Apple IIs with 48 KB.The 5-by-7 grid was too small to achieve bare-minimum legibility of Chinese characters. Engineers had to change the size of the Latin alphabetic grid to make it legible for Chinese characters. It was now larger than the 5-by-7 pixels. For the most common Chinese characters (either in simplified or traditional forms, but not both) the memory needed to store the bitmaps would be approximately 256KB. This is four times the capacity of most personal computers on the market in the 1980s. This is even before you account for RAM requirements for the operating systems and the application software.This is the setting for one of the greatest engineering histories of modern computing. It's a story of engineering ingenuity and entrepreneurial daring that offers a rare look at the global development of digital revolution.This article is the first in a series of TechCrunch articles. I will be examining the Sinotype III, an experimental computer that handled Chinese-language inputs and outputs. The Sinotype III was built on top of an Apple II, but equipped with a custom-programmed word process and operating system. It served as a proof-of-concept that showed how Western-made computers could be translated into Chinese.This is the first of two parts. I will discuss the technical difficulties that computer memory, fonts, and operating systems presented to the creators. And how they came up with innovative solutions.A newly graduated graduate with no job prospects immediately shows her chutzpahOur story starts with the Graphic Arts Research Foundation, the organization that arguably gave birth to Chinese computing. Samuel Hawks Caldwell, an MIT electrical engineer, invented the Ideographic Composing Machine (also known as the Sinotype) in the 1950s with GARF funding. The project was stopped in 1960 after Caldwell's untimely passing. The Sinotype project was maintained alive through the 1960s and 1970s by a variety of parties including RCA, Itek Corporation and, finally, GARF.Louis Rosenblum was a key factor in Sinotypes' homecoming. He was born in New York City in 1921. He graduated in 1942 with an undergraduate degree (Applied Math) and was a member of the MIT community. After studying under Harold Edgerton (the world-famous professor of electrical engineering) Rosenblum began work at Polaroid in 1942. He worked with Edwin Land on a number of projects, including instant photography. He moved to Photon in 1954 and worked on photocomposition using non-Latin writing methods. Rosenblum was deeply familiar with Caldwell's pioneering work in Sinotype. He adopted the project and revived it when GARF hired him as a consultant in mid-1970s.GARF worked on the Sinotype project until the 1980s. By then, it had established an advisory board that included renowned scholars as well as people with extensive China experience. Susumo Kuno, a Harvard linguist, joined the board. Richard Solomon, who is well-known for his crucial role in Richard Nixon's 1972 visit to the PRC and was then the head of the Social Science Department at RAND Corporation, also joined the board.This brain trust was impressive, but GARF's breakthrough on the Sinotype Project was the jump from a minicomputer-based system, Sinotype II, to one based upon a microcomputer, Sinotype III. It was initiated by a college student who had only worked for two weeks in data management for the Sinotype II project, which was completed in 1979. Bruce Rosenblum was Louis Rosenblum's son.Bruce, a University of Pennsylvania undergraduate and aspiring photojournalist was trying to balance his time between his coursework and his role in the photo editor of the student-run independent newspaper Daily Pennsylvanian. It was remarkable in its use of equipment and the depth of expertise of the students who ran it.Bruce's junior year saw the paper's typesetting equipment (two Compugraphic typesetters), reach the end of their lives and had to be replaced. Bruce, along with three of his student colleagues, helped to research replacements and settled on a $125,000 contract with Compugraphic in Wilmington, Massachusetts.Bruce knew a lot about the Sinotype project, thanks to his father. However, he was not involved in it. A pivotal moment occurred in May 1981. Bruce was just finishing his final exams and stopped by the office of the paper. Eric Jacobs, his colleague, was present and working on a TRS-80 Model II personalcomputer from Radio Shack. Jacobs was pondering the possibilities of using this microcomputer to manage the newspaper's business operations. Bruce stayed there for about 30 minutes before moving on to his day.He remained captivated by those 30 minutes, however. Bruce shared his experience with me via email. It was the first time he had ever seen someone work on a microcomputer. These few minutes inspired the entire Sinotype III project, and ultimately my career in computer engineering.Bruce made an off-the-cuff comment in a telephone call later that week to his father. Bruce made a comment about the high cost of the Data General hardware GARF was using to build Sinotype II. He said that someone could program something comparable or better with a microcomputer at a fraction the cost, perhaps as low as $10,000, as opposed to the over $100,000 GARF was funding.His father was fascinated. Louis asked Bruce if it was possible for him to program such a machine. Bruce had no formal computer science training, but he was an avid user of computers since high school. He also learned BASIC and PDP-8 assembly languages. He did respond to his father's question with the chutzpah and wit of a freshly graduated graduate without any immediate job prospects.Bruce met with Bill Garth and Prescott Low in New York to present his Sinotype III proposal. Bruce wore a three-piece suit to the meeting. Bruce's formal proposal cited $7,500 in hardware and $5,000 for programming costs. A Chinese word processor running on an Apple II was promised, and it would be delivered within four months. This would lower the cost of such an Apple II-based machine by at least one order of magnitude if it worked.Bruce was offered the job. He continued to program Sinotype III between June and November 1981. This allowed him to balance his time between his full-time job as tour guide at Independence Hall, Philadelphia, and his full-time job with the National Park Service. He would hand-write assembly code during daytime breaks and then transcribe it at night. Bruce's job as a tour guide ended in 1981, so he spent two months working on the code, and finally delivered it to GARF.Memory hackingComputer memory was the first issue that GARF and Rosenblums had to face. In an effort to squeeze as much memory out of their early Chinese computers, developers explored all options. Two strategies will be discussed, sometimes used in isolation but often together: Adaptive Memory, and Chinese Character Cards.The Sinotype III system consisted of five components: a 12-inch Sanyo DM5012CM monitor; an Epson MX70 printer; a Corvus 10-MB Rigid Disk Storage to store the Chinese character bitmap data and the corresponding descriptor codes; and an Apple Disk Drive to store text files.The Apple II was shipped with 32 KB RAM. It can be extended to 48 KB via the motherboard. Bruce Rosenblum, an email recipient to me, noted that we had reached the maximum amount of RAM before the Apple II was even available in the store. However, 48 KB was not enough memory for Bruce's purposes. Bruce decided to make a standard modification that was used by power users of the time: to insert a 16 KB card into slot 0. This would bring the total memory available to 64 KB.However, even this was insufficient. He said that I needed more RAM to store the full encoding system and the 16-by-16 bitmaps of the 100 most common ideographs.He started to look into a mod for the Apple II that not many had ever tried. He said that he had found a way to put a 16 KB second board in slot 2. This gave him a total of 80KB. He said that although it was not standard, it worked with components from the store.However, this modification enabled the machine to surpass its limitations. The Apple II's 6502 microprocessor could only access 64 KB of memory. This meant that even with Bruce's 16 KB bootstrapping in with the second memory card, there was no way for the Apple II and other Apple II computers to access additional memory addresses simultaneously. This mod was so unusual that the Apple rep was stunned when he mentioned it to an Apple engineer during one of his many conversations.Bruce wrote his own assembly language program to allow the Apple II access to 80 KB of memory instead of 64 KB. His custom-designed program allowed him to select between two 16 KB banks. This meant that although 64 KB of memory would be available at once, Bruce could quickly switch between the memory expansion cards and trick the computer into accessing them at speeds that were negligible from his perspective. This allowed for the addition of up to 400 Chinese characters into the on-board memory, which resulted in 25% more system memory.Bruce gave the final code to GARF a week before Thanksgiving. He then embarked on a backpacking trip that would take him around Europe and Asia. The development of Sinotype III was left to Louis Rosenblum and GARF. Bruce, however, continued to be a consultant and exchanged frequent correspondence from anywhere in Europe, China or India he was at the time.Moving towards real-time Chinese typingBruce and Louis estimated that only 600-1000 Chinese characters could fit into the onboard memory, despite their clever mod. The size of Sinotype IIIs operating systems, program applications, and the memory requirements for each Chinese character mean that the vast majority the Chinese characters in the machine lexicon will need to be stored elsewhere, either on floppy discs or an external hard drive.Bruce briefly considered PROM (programmable-read-only memory) chips early on. However, this idea quickly proved to be dead. The largest PROM chips available on the market at the time were 2 KB. This translated to just 28 to 51 Chinese characters. Bruce would need either 138 or 250 PROM chip to store the 7,000 characters. He noted that this was a lot of chips.Bruce considered the possibility to store characters on floppy discs. This too proved infeasible due to the number of disks required and the slow access/retrieval speeds involved in retrieving bitmaps of character bits from floppy drives storage. GARF chose to equip Sinotype III with an external drive. This was a rare accessory for microcomputers at that time. GARF decided to store thousands of lower-frequency Chinese characters on-site, in order to overcome its memory limitations. This was a Corvus Rigid Storage 10 MB.However, this had negative consequences for Sinotype III's operating speed. Hard drives are cumbersome in the computing space-time continuum, where most operations occur at subsecond speeds. They relied on magnetic platters platters which rotated inside the device, similar to a record player, in particular at this time. A head read the contents of different tracks, much like a needle reading the grooves on a record. The head's location and the rotational position of each disk at the time of retrieval were factors that influenced the speed of data retrieval. It was like waiting for the bus to return to the stop, only to discover that it had just left.The retrieval time for Chinese characters stored on hard drives was 10 times slower that those in RAM. The retrieval time of Chinese characters stored on RAM was approximately 100 milliseconds per Chinese character, a unit of time that is not easily perceived by human cognition. For characters stored in external storage, however the input of any character required more than a second to access and retrieve an entire unit of time, well within the threshold for human perception.In the context of mid-1980s personal computers, users were rapidly becoming used to real-time typing and a one-second input speed would have been devastatingly slow. Additionally, a second is 10 times longer than 100 milliseconds. This means that an average user would feel the difference every time they wanted to input lower-frequency characters.Louis Rosenblum came up with an idea to solve this problem: adaptive temporary storage. This would allow the user to change the number of characters in RAM based on what they have just typed. On initial boot, the on-board RAM of Sinotype III would only be equipped with a predetermined number of high-frequency characters. As noted above, it would take approximately one second to input any hard-drive-based infrequent characters. As each of the less common ideographs are keyboarded, he will explain in a letter that the code and dot matrix patterns would be noted in the random memory. This would reduce retrieval times.Chinese-on-a ChipThere were still many thousands of characters beyond what could be done with adaptive memory and toggling. Although high-frequency Chinese characters made up a significant portion of the overall usage, any type of technical or specialist content would have brought the user back to the off-site repository. If Chinese computing is to offer the same instantaneity as English-language counterparts, it would be necessary to bring more of these low-frequency characters on-site.In the late 1970s and early 80s engineers began to look at a new hardware solution. These were called Chinese Character Cards (Hanka), Chinese Cards(Zhongwenka), Chinese Character Generators (Hanzi Zimo fashengqi), Chinese Font Generators (Hanzi Zimo fashengqi), or Chinese Character Generators (Hanzi zimo fashengqi). These cards had thousands of Chinese input encodings and bitmaps hardwired to them. They served the same purpose as an external hard disk, but with much faster speeds and more reliable performance.Research at GARF was not centered on Chinese-on-achip cards. They were actually a result of earlier custom-designed Chinese systems that existed before the advent of personal computing. These systems included the Ideographix IPX by Chan Yeh and the Olympia 1011 which were equipped with microprocessors that only had one purpose: to generate character bitmaps and store input descriptors. The Olympia 1011 Chinese word processor was a single-purpose electric Chinese typewriter. One of three Intel 8085 processors was devoted to Chinese character generation.These character generators became commodities in the 1980s and were made into products. To access this type of on-board generator, one no longer needed to purchase a word processor like the Olympia 1011. One could instead purchase a Chinese Character Card, and then install it onto the computer of one's choice.Tsinghua University was one of the first centers to specialize in Chinese computing. It developed a card that could store approximately 6,000 Chinese bitmap patterns, 32-by32 dot matrix format. There were many Hankas on the market by the late 1980s. They were manufactured and sold by companies from Japan, China and Taiwan as well as the United States.The Chinese-on-achip approach was so popular and widespread that almost all computers with Japanese or Chinese language capabilities had a character generator card.From the 1950s with Caldwells Sinotype, to the father-son Rosenblum team in the 1980s and GARF around Sinotype III by the 1990s, solving memory problems related to Chinese characters was the key to opening China's market to computing. Hacking computers with more memory, developing adaptive memory algorithms to prioritize characters and building dedicated hardware were some of the key factors in the Chinese computer revolution.The next step was to connect the computer to all other devices. TechCrunch will be presenting part 2 of this series. It will discuss the difficulties of programming and designing early computers that can handle Chinese text output.