Key among the many technologies that drive the IT industry are semiconductors. This is the base technology that enables all of the other parts of the technology business. Whenever semiconductor technology hits a speed bump, all of IT bounces. Products based on semi technology are migrating into all aspects of life, mostly to our benefit. Other basic technologies are optical for communications and MEMS for micro sized electonic actuators and sensors.
For the last few years it looked like semi technology was seeing the end of our ability to advance to finer resolutions which means more devices in a given space and lower power (and cost) per device. There were serious problems in Lithography caused by the inability to optically project finer masks onto the photosensitive resist to define the patterns on the chips.
Most of these obstacles have disappeared or been mastered through substantial research over the last year. Recently, the technology to enable building the .065 micron class has reached feasibility. [ref] Semiconductor specialists expect to start building .065 micron lines by 2004. Since we have just moved into 0.13 production and .09 micron has just delivered preproduction samples, there is plenty of time to get .065 ready.
The costs of semiconductor fabrication (fabs) increased by large amounts with each new advance. Once unthinkable, billion dollar fabs are common, and the new .09 and .065 fabs are likely to exceed two billion dollars. Designing and building new chips has became a major investment in time and money, which limits the newest technology to those chips which have enough volume to be affordable after all costs are amortized.
Along the way, new discoveries in materials and processing have broadened the number of options available to build new devices. One such material, Silicon Germanium (SiGe), has been around for a while but recent processing advances for SiGe and frequency limits in standard CMOS silicon have made SiGe first choice for the highest frequency applications.
Recent advances to Silicon Germanium Carbon (SiGe:C) have extended the frequency range to 80 GHz in 0.35 micron, with over 200 GHz expected at the 0.18 micron technology level early in 2003.
These advances include digital radios and gigahertz amplifiers used in cellphone towers. Better amplifiers means fewer towers to cover a given area, dropping costs and contributing to the rapid uptake of cell phones. These advances will join other technology advances that enable the next set of new industries.
Recent advances in processors have insured that we will have plenty of power for current PC applications. It is the lack of must have new applications that demand higher performance that contributes to the current malaise of the personal computer and IT industry. That cause should vanish in the next year.
Another change in processors is the development of a wider range of mass produced processors. From the System On a Chip (SOC) with everything needed for a system except the actuators to multi-processor chips that require special cooling for highest performance. New chip designs include the Cell design from IBM and Sony, hyperthreading, and high end multiple core chips available in low and mid-range servers.
For companies and the upcoming computer utilities, smaller and lower power chips built in Blade systems will deliver more processing power at lower cost in a smaller space. The recent 1999 to 2001 rapid performance increase of desktop processors [ref] will repeat over the 2003 to 2005 timeframe and continue stirring the imaginations of inventors and entrapranuers.
Recent announcements about Blade development [IBM's big blades] have made it clear that this will become the preferred way to aggregrate large amounts of computing capability. Current technology, using six or seven foot tall industrial racks can contain forty single or dual processor systems in 1U sizes.
Blades can contain 14 or more dual processors in a 7U system, doubling or quadrupling the capability in the same space. Just as important, the blade systems internalize the communications links, running at higher speeds and eliminating messy and hard to manage cabling.
The result is that small and medium sized companies with 100 to 1000 employees can put all their required computing capacity in the equivalent of a large closet. Increasing capacity is simple - just purchase a blade or two, plug them in and capacity is on line. The economics of this approach will rapidly clear most alternatives from the market.
CPU designs are also about to evolve to a new level. I call it evolution because it continues a process that has been happening right in front of us for decades. It is the next step of building more processors in a given space.
Here's the development sequence:
The cell chip is a natural step in the process of building more processor power in a given space, then shrinking the space. I think there are at least two more steps:
Based on the evolution of systems listed above, Blades will soon replace most multiprocessor systems in rack mount enclosures. The blade systems offer more processing and communications capacity, fewer resources (power, A/C, floor space, cables), easier expansion and management. It's no contest.
Another consequence of this evolution is that Intel's big bet (about $5 billion) on the Itanium will have very limited application and success. It will never become a mass produced chip like Xeon because of its size and power requirements, even when shrunk by smaller process technology.
The problem Intel faces is simple. In the light of the evolution of systems, Intel took a wrong turn at the conceptual design level by choosing one large interconnected processor. The smaller cell designs will win because they are easier to design and build. They will also be less expensive to produce, easier to program, and more flexible in application.
Had Intel delivered the Itanium close to the initial schedule in 1998, it's success would have been very likely. The late delivery, which continues to slip, has allowed competitors to deliver improved processors. New design concepts like the cell have arrived to provide better ways to deliver high performance.
Only the sale of Compaq's Alpha and HP's commitment to phase out the PA-RISC 8000 series processors gives the Itanium any chance to succeed. Intel still faces two powerful opponents with mature processor lines and massive software libraries: IBM and Sun.
Sun SPARC processors have fallen behind in the performance race, but Sun will respond with faster processors and leverage their large systems designs to stay competitive. IBM will enhance it's Power4 line and implement Cell based chips. I expect IBM will stay at the top of the performance list for some time.
We have had welding and entertainment robots robots for decades, toy robots for several years, but useful robots for the home? Sounds like a bit of Science Fiction (SF), doesn't it?
It isn't SF any longer. Like the Intel 4004 in 1972, this robot mini vacuum cleaner, Roomba, is just a hint of a leading edge that will become a new industry in the 2003-2005 period. These simple robots will be controlled by automotive class embedded SOC processors. Robots for more complex tasks will require more processor power, likely to be spread across multiple SOC chips, customized for specific tasks. The evolution of robot chips will emulate the development of automotive chips, at a faster pace.
Most home robots will tie into the home wireless network. Bluetooth, the most power and cost effective technology for this purpose, will be the link. Low power short range wireless will enable instructions to be sent and status to be received at a personal computer. Part of the home PC software will monitor the robots, schedule operation and alert the family to unusual situations.
Along with robots, the wired home will take over security monitoring, use voice recognition to allow access to the home or facilities, provide water and energy conservation, control the environment and serve as the central store for media and entertainment. It will require a better class of PC components than the current IDE drives and a better OS than is usual for desktops, with voice activation standard.
Although some of this is still SF, it is no longer very far off. I expect that the dishwashing robot will be the hardest program to develop. However, walking robots were once an absolute roadblock, but now you can read about and see them in action . So I expect the dishwasher robot challenge will be solved too. For a SF look ahead on robots, read Robert Heinlein's "Door Into Summer," written almost 50 years ago.
One of the essential pieces of home infrastructure is the proliferation of wireless capabilities. Now that the range of capabilities is close to complete, software integration can begin. Development has been so rapid in this space that some early standards are already surpassed by those only 18 months later.
From Bluetooth in devices to 803.11x in business and home to Ultra WideBand (UWB) for multimedia and entertainment or bandwidth demanding local applications, there is now a technology to meet any need. It isn't all deliverable yet, but the basic pieces are in place.
The next step is robot software, and this will require standards for robot functional interfaces to be developed. The actual wireless transmissions are not the issue, but the format and content of status reports and commands will be.
Delivery of a wide range of devices will depend on the robot industry developing a standards group like or within the W3C (World Wide Web Consortium). One quick way to develop these functional interfaces is for the vendors to adopt XML as a base for the standards, and rapidly develop prototype DTDs that specify the interfaces.
With some care, this approach can be made upgradable in the field so that devices may evolve over time via software upgrades. Some of that care must be applied to make it impossible for hackers to insert different DTDs to change operations. Safety will be needed once robots are in the home.
A safety monitor in ROM, or a separate custom SOC, that watches the results of the commands and prevents dangerous actions is one solution to this. Developing such a monitor for different classes of robots and qualifying it for safety with specific models will be the basis for a new industry.
A third component of today's web is software standards. Software development has always been a difficult process because of the level of detail and complexity required to complete a program. Standards that define how components interface have been around a while, but they used to be low level APIs and differed from vendor to vendor.
Only in the 1990s did the development of higher level standards that were universally accepted become the norm. Not only could developers avoid having to once again reinvent a special wheel to fit a unique hexagonal axle, the axle interface was now operational, avoiding lengthy debugging delays.
Software has finally overcome some of the barriers that have existed since the 1960s. The components of this level jump have developed from:
The key results are better productivity and multiple OS support, with source code available for fixes and improvements from anyone. Productivity improvements come in part from eliminating dependence on C or C++ for all minor and major software development. The proliferation and wide use of scripting languages is a programming capability I have wanted since the 1980s, albeit in a form I did not expect.
The net result of this web of changes is something more than the Internet. It is that building powerful services and capabilities of all types is not only feasible, but faster and less expensive than as recently as 2000. Most of the new standards have been ratified and put into immediate use since that time.
While the web is the most obvious beneficiary of this wide range of improvements, both personal programming and tools have exploded as individuals build the tools that they want. One class of tool which is just coming into its own is the Platform, a program which supplies a basic infrastructure for a specific application area which is customized by plugin tools.
Eclipse is a good example of the platform class of tools. These are frameworks that provide a structure where special purpose components can plug in and share work together without any specific platform programming other than observance of the component interfaces. The platform is then customized for an individual or group by plugin components that provide specific functions and services.
Eclipse was developed by IBM and donated to Open Source after a reported $40 million was spent developing it. Another platform, Zope, a web server on steroids with a very wide range of plugin extensions. Developed in Python by the Zope Corporation, Zope is Open Source, and the company will sell services to support its use commercially.
Eclipse, Zope, Apache and similar projects represent the leading edge of new application software development. Building good tools and applications are a demanding process, beyond all but the largest corporations. With the power of the Web and Open Source coupled with commercial backing, these past limitations are disappearing. Developments like these platforms, based on open standards, will become a major force in future software development.
These current platforms represent a tool class that will soon include other application areas like accounting, data mining and visualization. All platforms will bring a common interface to a class of applications that have similar objectives but require special or customized processing for some parts of the job.
The next step in software development will be a Software Factory, where a library of tools and components will be searchable by type and functionality. Programming will become an industrial process of development by selecting components from the library and adding the User Interface for the finished product.
These software factories will create a new type of software - Built On Demand. (Ref: Earlier column about BOD). They will drive the skills needed to design applications to higher level conceptual design integrated with overall company objectives. Factories will broaden the range of special applications, one time or infrequent use, that would not be economical with current practices.
Programming will still be a challenge, but on a different level. Bit twiddlers, necessary in the early days, have a limited future in hardware driver code. This area too will expand slowly as more chips are produced, constrained by hardware interfaces moving to standards, reducing demand for custom drivers.
The final piece of Waves of Change will cover the following areas:
[30]