The Revolution Of High Speed Processor For The Masses

The prevalence of various types of electronic devices is rising rapidly in recent years. Almost every device contains a CPU (central processing unit) to execute most instructions. It is a chip of integrated circuits with many components only several hundred micro-meters square in size. Consumers desire products which are power-saving, high-speed and reliable. Engineers are attempting varies ways to improve processors to meet users' demand. However, the development of consumer processor faces bottlenecks.

Users are eager for high-performance of electronic devices. This is the main reason for enhancing computation capability of CPUs. Processors launched within last 5 years allow most present operating systems to run smoothly, and execute frequently-used programmes without evident delay. It is no longer the case when users want to play videos or games. High-definition videos adopt new compression standard in order to maintain small file sizes. H.265 is a good example. Analysis shows that whilst it improves video quality, the compression ratio of this new standard is twice than that of H.264 (the previous standard) [1]. H.265 increases computational complexity. Besides, game development companies do not concern the hardware much. Many recently revealed 3D games cannot be run on most users' computers. ARMA 3, for example, requires at least a dual-core 2.4 GHz, but a quad-core is officially recommended [2]. Generally, the quality of user experience is heavily determined by latency (time delay) and energy consumption of their devices. It is advisable to maintain both the latency and power low in consumer processors.

Dual-core and quad-core CPUs are examples of multicore processors which are the latest practical approach to fastening data processing. Before this, experts pursued some methods like raising the minimizing transistors and clock frequency.

Fast microprocessor can be obtained by increasing the transistor density. The main element of a processor is complementary metal-oxide-semiconductor (CMOS). Manufacturing technology of consumer processors has been shrunk from 32nm to 22nm. These numbers refer to the gate length from source to drain in one single transistor, as Figure 1 shows. In scaled down transistors, gate oxide thickness is 1-2nm or even less. In this case, electrons and can easily flow through gate oxide without control, leakage current occurs [3]. This phenomenon is described as tunnelling in quantum mechanics. As a result of tunnelling, accuracy of data processing drops and power consumption increases. Chips become inefficient.

Figure 1: Schematic cross-section of a CMOS transistor, adapted from reference 3.

In order to interpret instructions, a CPU needs a clock line generating pulse at a certain frequency. High frequencies allow faster computation. Nevertheless, simply enhancing the clock frequency does not mean better computing performance. It relies on the progress of manufacturing technology which is facing a technological barrier. At a high clock frequency, the interconnections among transistors have to be redesigned and compact, so that the processor can maintain the accuracy and reliability. To some extent, a properly increased clock frequency will obtain a higher computing ability. Few worldwide experts are working on overclocking. They force computers to run over the recommended clock frequency. CPU frequency is raised up to 8.709 GHz in June 2012 [4]. However, it is hazardous since any fault in the design of an overclocking system will cause an explosion and have potential personal injury.

Also, there is enormous financial investment on researching powerful nano-meter size transistors. Intel, one famous processor manufacturer, has been spending 10 billion dollars on researching new manufacturing technology annually from 2009 to 2012 [5]. Actually, manufacturers may not receive equivalent returns in the future.

Design complexity and unreasonable costs of production limit the performance of single-core processing unit. The singular focus on enhancing the existing structure of CPU could be an obstacle of the industry. The huge market of high-speed and energy-efficient devices drives researchers to think more outside the box. Fortunately, meanwhile, multi-processor architecture is proposed as a new concept.

The chip multi-processor (CMP) architecture is a combination of single cores (known as sub-processors) which operate in parallel. In principle, parallelism in CMPs is able to process more instructions in the same amount of time. The cores are isolated but still have communication between each other. There is no theoretical difficulty in production. In the early years of 21st century, many different prototypes emerged, e.g., Hydra [6], Raw [7], TRIPS [8]. Two semiconductor chip tycoons, Intel and AMD, revealed a number of consumer multi-core processors from the year 2005. It seems possible to completely replace single core with CMP.

Multicore has advantages. Despite the relative simplicity of design, it can also be produced as a combination of different types of cores. Each core has specific strengths. This architecture utilizes the superiority of individual cores. For instance, a multi-processor system is able to decode different video frames and graph pixels at the same time, or to run different software simultaneously. Furthermore, manufacturers can invest less money on researching and development, which reduces the cost of production and product prices. This new concept can be deemed as a milestone in CPU industry.

However, multicore is not almighty. To get the benefits of CMPs, software must be able to recognize it. Single-threaded applications will not run any faster on these architectures. Writing programmes working efficiently on a multi-processor system is not easy as new types of bugs may occur. Thus, the complexity of computer software is one negative aspect and cannot be avoided until now. Moreover, communication between cores becomes excessively frequent. This requires high-bandwidth. Improper connections between the cores and other units (e.g. storage devices) inside a processor will cause the latency of network to rise. In addition, tasks must be delivered clearly. Otherwise a multicore processor will spend all time on communication and no work will be done [9].

Several solutions have been introduced. Some programming languages (e.g., Go [10], Rust [11], Erlang [12]) are launched to reduce the complication of computing in concurrency. Additionally, multi-core systems are categorized into homogeneous (contains identical cores) and heterogeneous (contains diverse cores). A well-designed heterogeneous multi-core system could avoid the problem caused by high-bandwidth. In this case, each core has distinctive tasks and do not need to exchange much information between each other.

Designers and scientists are researching on the substitutes of existing processors. The flat form of graphene, made of carbon, performs like a semiconductor. Higher performance of processors becomes more realistic if graphene is used in a transistor to connect source and drain [13]. Besides, other researchers have already made progress on quantum computing, DNA computing, spin wave devices, and more.

With the prevalence of fast CPUs, an ethical issue becomes a focal point. Modern encryption technology is established on unsolvable problems in mathematics and it will spend centuries to decipher a code by using brute force attack (based on high speed computation). In fact, high performance computers among the public allow crackers to use distributed computing to obtain plaintexts, enciphered messages, within a short period of time. Businesses and governments classified information will not be safe anymore.

The revolution of consumer microprocessors is in progress. Designers used to donate huge efforts on diminishing the size of transistors or raising the clock frequency. These approaches involve trade-offs, e.g., increased power consumption, reduced reliability. Multi-processor architecture is a better solution, although it still faces some new problems. There is no master key in engineering. Researchers need more divergent thinking to improve the processing units. In general, faster devices do not only meet the desire of users, but also boost the efficiency of modern society in whole.