The materials science of semiconductors 9百二 sa310 uesaJoydsou ou-g39 8 9:4: 喜 8 n-o 8 eMsueun)Au明 曾喜了 electrons relax back onto the Er atoms. When a light pulse passes through the doped region it is amplified by gain involving the Er atom defect states The above are only a tiny fraction of the many remarkable discoveries and innovations that have sustained the recent explosive growth in microelectronics
6 electrons relax back onto the Er atoms. When a light pulse passes through the doped region it is amplified by gain involving the Er atom defect states. The above are only a tiny fraction of the many remarkable discoveries and innovations that have sustained the recent explosive growth in microelectronics. Figure 1.1: Shows the progress of light-emitting devices as a function of time. Note that all technologies are converging to that emits the visible light. Figure based on data in Sheats [1] with additional data from Forrest [2], Andrade and Brown [3] the efficiency of flourescent lamps b and Cree Research press release [4]. ecause the recent devices use a high-energy emitter to drive a phosphorescent material The Materials Science of Semiconductors
An Environment of Challenges Developments in both electronic and optical devices continue at a staggering rate Most of these are based on the fundamental science of the materials and processing methods described here and in other texts 1. 3 AN ISSUE OF SCALE As noted above, electronic devices have improved remarkably in recent years. In 965, Gordon Moore observed that the number of transistors per square inch had doubled every year since the integrated circuit was invented in 1958. His projection that this trend would continue has come to be known as moore's law and it has hung like the Sword of Damocles over the minds of most microelectronics engineers who wonder how long they can maintain it. Despite many firm predictions of its imminent end, Moores Law has proven correct for over 40 years, as shown in Figure 1. 2. The early days of integrated circuits saw only a few devices in the circuit. The number of devices grew to hundreds in the 1960s, thousands to tens nousands in the 1970s to the current numbers of millions to tens of millions. The doubling time for transistors per device has been closer to 18 months than one year since the statement of the Law, although recent gains have, remarkably, been faster There is no obvious reason why it should not be possible to continue the Law until devices shrink well below the tenth-micron scale(individual devices with critical lengths less than 10 cm). Beyond that it is hard to see how the trend can continue, and a new device paradigm will be needed. Multilayer structures are possible even based on current technology and whole new concepts in device fabrication are anticipated, although none has yet shown a likelihood of replacing the conventional structures Along with this staggering increase in device density, performance of the devices has improved dramatically while the price has remained roughly constant, fueling a vigorous market for upgraded computers. To maintain this market and encourage people to purchase new computers, it is necessary for each generation of circuitry to mprove sufficiently to justify upgrading hardware. The inevitable failure to improve performance will presumably change the entire nature of microelectronics from a research and development driven field to a commodity driven field. It is our hallenge to both stave-off this change as long as possible based on the curren levice designs and to create new technology fuel for the microelectronics industry based on concepts not yet conceived The enormous improvements in device performance have come at the cost of intense and ongoing research and development efforts. These have resulted in creation of new materials and processes, and dramatic improvements in our understanding of the materials involved. There is no doubt that electronic grade Si single crystals are the lost-perfect and most highly studied materials ever produced. Further improvements are becoming harder to achieve and the challenges faced are becoming more fundamental
An Environment of Challenges 7 Most of these are based on the fundamental science of the materials and processing methods described here and in other texts. 1.3 AN ISSUE OF SCALE As noted above, electronic devices have improved remarkably in recent years. In 1965, Gordon Moore observed that the number of transistors per square inch had doubled every year since the integrated circuit was invented in 1958. His projection that this trend would continue has come to be known as Moore’s Law and it has hung like the Sword of Damocles over the minds of most microelectronics engineers, who wonder how long they can maintain it. Despite many firm predictions of its imminent end, Moore’s Law has proven correct for over 40 years, as shown in Figure 1.2. The early days of integrated circuits saw only a few devices in the circuit. The number of devices grew to hundreds in the 1960’s, thousands to tens of thousands in the 1970’s, to the current numbers of millions to tens of millions. The doubling time for transistors per device has been closer to 18 months than one year since the statement of the Law, although recent gains have, remarkably, been faster. There is no obvious reason why it should not be possible to continue the Law until devices shrink well below the tenth-micron scale (individual devices with critical lengths less than 10-5 cm). Beyond that it is hard to see how the trend can continue, and a new device paradigm will be needed. Multilayer structures are possible even based on current technology and whole new concepts in device fabrication are anticipated, although none has yet shown a likelihood of replacing the conventional structures. Along with this staggering increase in device density, performance of the devices has improved dramatically while the price has remained roughly constant, fueling a vigorous market for upgraded computers. To maintain this market and encourage people to purchase new computers, it is necessary for each generation of circuitry to improve sufficiently to justify upgrading hardware. The inevitable failure to improve performance will presumably change the entire nature of microelectronics from a research and development driven field to a commodity driven field. It is our challenge to both stave-off this change as long as possible based on the current device designs and to create new technology fuel for the microelectronics industry based on concepts not yet conceived. The enormous improvements in device performance have come at the cost of intense and ongoing research and development efforts. These have resulted in creation of new materials and processes, and dramatic improvements in our understanding of the materials involved. There is no doubt that electronic grade Si single crystals are the most-perfect and most highly studied materials ever produced. Further improvements are becoming harder to achieve and the challenges faced are becoming more fundamental. Developments in both electronic and optical devices continue at a staggering rate
The materials science of semiconductors E 107 Moore's Law Selected Circuit Data y10 o∽c 10 10 10· 195019601970198019902000 Year Figure 1.2: Shows the trend known as Moores Law for microprocessor circuit density. The minimum feature size has decreased proportionate to the square root of the circuit density. Each plotted point corresponds to a marketed product. Figure based on data on the Intel" web sitehttp://www.intel.com/technology/mooreslaw/index.htm20061 Advanced production device dimensions have now shrunk to roughly 0.1 um minimum feature sizes(see Figure 1.3). The challenges of producing a device of this scale become more obvious when one considers doping of the semiconductor and the electron current density passing through the device. a transistor with a control region length and width of -0.1 um(10 cm)and a thickness of 50 nm(5x10 cm)is current technology. At this scale, the control volume is 5x10 cm. The atom density of silicon is 5x10 atoms cm. This suggests that the critical volume of current transistors within which the device is switched on or off contains only million atoms Doping the semiconductor with one part per million impurity atoms(5x10 cm),a ypical value for older devices, means the control region would contain only 25 mpurity atoms. The removal of a single dopant atom would therefore correspond to a 4% change in doping level. Such doping levels do not provide adequate conductivity or reproducibility. Consequently, small devices have doping levels closer to 1x10cm or 0.02%(at or near the solubility limit for the dopant).Even at
8 Figure 1.2: Shows the trend known as Moore’s Law for microprocessor circuit density. The minimum feature size has decreased proportionate to the square root of the circuit density. Each plotted point corresponds to a marketed product. [Figure based on data on the Intel® web Advanced production device dimensions have now shrunk to roughly 0.1 µm minimum feature sizes (see Figure 1.3). The challenges of producing a device of this scale become more obvious when one considers doping of the semiconductor and the electron current density passing through the device. A transistor with a control region length and width of ~ 0.1 µm (10-5 cm) and a thickness of 50 nm (5x10-6 cm) is current technology. At this scale, the control volume is 5x10-16 cm3 . The atom density of silicon is 5x1022 atoms cm-3. This suggests that the critical volume of current transistors within which the device is switched on or off contains only 25 million atoms. Doping the semiconductor with one part per million impurity atoms (5x1016 cm-3), a typical value for older devices, means the control region would contain only 25 impurity atoms. The removal of a single dopant atom would therefore correspond to a 4% change in doping level. Such doping levels do not provide adequate conductivity or reproducibility. Consequently, small devices have doping levels closer to 1x1019 cm-3 or 0.02% (at or near the solubility limit for the dopant). Even at site: http://www.intel.com/technology/mooreslaw/index.htm, 2006.] The Materials Science of Semiconductors
An Environment of Challenges this doping level the control volume contains only 5000 dopant atoms. Reduction in scale of another factor of five in the lateral dimensions is easily foreseen in current echnology. With no reduction of thickness, the control volume would contain only 1000 dopant atoms at the higher doping level. A variation of only 10 dopant atoms well within what one might expect for fluctuations in doping sources) corresponds to a 1% change in doping level. Such changes can affect the resulting performance Device designs must therefore be more capable of accommodating materials variability. The challenge will grow further if the smallest transistors to date with control volumes of the scale of 2x10"cm(10,000 atoms)are to be manufactured Data in integrated circuits is carried by small bursts of electrons. Typical devices ow operate at >10 Hz(1 GHz) with some as high as 5x10 Hz Current flowing for the corresponding cycle time(10" s)transfers only -2x10 electrons per amp of current.a current of i na transfers only one or two electrons in a nanosecond while single electron"transistors have been produced that can be turned on and off with single electrons, devices responding to such small amounts of charge are not yet practical. A more reasonable number of electrons to activate a device or store a data 10μAat1Gh transfers 80.000 100000Acm ectrons per half cycle Gate Channel 0.1pm 1 part per million doping=25 atoms discussed here. The current densities(105 A cm)are sufficient to cause conductors to fields across the gate dielectric(5x10 V cm")are barely supportable by even the nearly transferred through the device is so small that noise becomes significant
An Environment of Challenges 9 this doping level the control volume contains only 5000 dopant atoms. Reduction in scale of another factor of five in the lateral dimensions is easily foreseen in current technology. With no reduction of thickness, the control volume would contain only 1000 dopant atoms at the higher doping level. A variation of only 10 dopant atoms (well within what one might expect for fluctuations in doping sources) corresponds to a 1% change in doping level. Such changes can affect the resulting performance. Device designs must therefore be more capable of accommodating materials variability. The challenge will grow further if the smallest transistors to date with control volumes of the scale of 2x10-18 cm-3 (10,000 atoms) are to be manufactured. Data in integrated circuits is carried by small bursts of electrons. Typical devices now operate at >109 Hz (1 GHz) with some as high as 5x1011 Hz. Current flowing for the corresponding cycle time (10-9 s) transfers only ~2x109 electrons per amp of current. A current of 1 nA transfers only one or two electrons in a nanosecond. While “single electron” transistors have been produced that can be turned on and off with single electrons, devices responding to such small amounts of charge are not yet practical. A more reasonable number of electrons to activate a device or store a data Figure 1.3: A schematic diagram of a state-of-the-art field-effect transistor such as that discussed here. The current densities (105 A cm-1) are sufficient to cause conductors to fail, the fields across the gate dielectric (5x106 V cm-1) are barely supportable by even the nearly perfect SiO2 gate dielectric, the number of dopant atoms in the channel limits the practical dopant concentrations to parts per thousand typically, and the total number of electrons transferred through the device is so small that noise becomes significant
10 The materials science of semiconductors value is of the order of a few thousand, corresponding to a required current of a few microamps in each cycle. A simple analysis of the 0. 1 micron device above would that a current of 10 HA(10-A)would correspond to a current density through the 5x10 cm- cross section of the device of 2x10- a cm. devices operating at 500 GHz require even higher current densities for practical operation, even though the number of carriers flowing in a cycle is reducec Current densities of this magnitude flowing through conducting wires produce an electron wind"with sufficient momentum to literally push atoms along the conductor in materials where atoms diffuse easily. This phenomenon of" electromigration"has een one of the long-standing causes of device failures. It is the primary reason driving the transition from aluminum to copper conductors in current generation devices and has been considered, in some cases, the major issue ultimately limitin the lifetime of operating integrated circuits At the same time that current densities are rising to unacceptable levels, threatening to melt devices during operation, and total atom numbers are forcing restrictions on the minimum doping levels in device control volumes, insulators are beginning to fail. The best insulator known is Sio2 grown by thermal oxidation of Si wafers. Such oxides can support fields of up to 10 million volts per cm. Do not depend upon air supporting such a field by placing your finger I cm from a 10 MV power source Even bulk SiO2 can not support such a field. These fields mean that a 1 V potential requires a minimum of I nm(10 cm) of oxide if there are no defects or thickness hanges present. In practical terms, more like 2 nm of oxide are required at this voltage. Shrinking the overall device dimensions has required shrinking the oxide dielectrics accordingly. Oxides are approaching I nm thick (-3 molecules). This has required reduction of the voltages, which produce dramatic changes in the design of switching transistors. The reduction in allowed voltages has spurred the development of new dielectric materials with higher dielectric constants Considering the issues above, one finds that even for the current technology devices, here are major restrictions facing electrical engineers that prevent optimization of the circuit performance. Current research is focusing on ways to alter the semiconductor to permit higher doping levels and faster motion of carriers through he device control volumes. Dielectrics are needed which will support the same electric fields as SiO2 while producing higher capacitance. Finally, conductors that n carry higher current densities without failure are under study. All of these issues must be overcome before the next generation of devices can be produced. To continue on Moore's Law. two such generations of device. with their incumbent changes in materials, will have to have been produced between the time it was written and the time you read this book. Imagine the exciting challenges that will face you
10 value is of the order of a few thousand, corresponding to a required current of a few microamps in each cycle. A simple analysis of the 0.1 micron device above would suggest that a current of 10 µA (10-5 A) would correspond to a current density through the 5x10-11 cm2 cross section of the device of 2x105 A cm-2. Devices operating at 500 GHz require even higher current densities for practical operation, even though the number of carriers flowing in a cycle is reduced. Current densities of this magnitude flowing through conducting wires produce an “electron wind” with sufficient momentum to literally push atoms along the conductor in materials where atoms diffuse easily. This phenomenon of “electromigration” has been one of the long-standing causes of device failures. It is the primary reason driving the transition from aluminum to copper conductors in current generation devices and has been considered, in some cases, the major issue ultimately limiting the lifetime of operating integrated circuits. At the same time that current densities are rising to unacceptable levels, threatening to melt devices during operation, and total atom numbers are forcing restrictions on the minimum doping levels in device control volumes, insulators are beginning to fail. The best insulator known is SiO2 grown by thermal oxidation of Si wafers. Such oxides can support fields of up to 10 million volts per cm. Do not depend upon air supporting such a field by placing your finger 1 cm from a 10 MV power source! Even bulk SiO2 can not support such a field. These fields mean that a 1 V potential requires a minimum of 1 nm (10-7 cm) of oxide if there are no defects or thickness changes present. In practical terms, more like 2 nm of oxide are required at this voltage. Shrinking the overall device dimensions has required shrinking the oxide dielectrics accordingly. Oxides are approaching 1 nm thick (~3 molecules). This has required reduction of the voltages, which produce dramatic changes in the design of switching transistors. The reduction in allowed voltages has spurred the development of new dielectric materials with higher dielectric constants. Considering the issues above, one finds that even for the current technology devices, there are major restrictions facing electrical engineers that prevent optimization of the circuit performance. Current research is focusing on ways to alter the semiconductor to permit higher doping levels and faster motion of carriers through the device control volumes. Dielectrics are needed which will support the same electric fields as SiO2 while producing higher capacitance. Finally, conductors that can carry higher current densities without failure are under study. All of these issues must be overcome before the next generation of devices can be produced. To continue on Moore’s Law, two such generations of device, with their incumbent changes in materials, will have to have been produced between the time it was written and the time you read this book. Imagine the exciting challenges that will face you! The Materials Science of Semiconductors