rus
In the third part of the article analyzes the perception of the industry community of Moore’s law in the past, present and future.
Теги: dennard’s law integrated circuits microelectronics microprocessor moore’s law закона деннарда закон мура интегральная схема микропроцессор микроэлектроника
We are turning to Moore’s Law again, but now from the perspective of emerging deviations that may result in denying its impact on micro-nanoelectronics development. This first came into sight with the need to limit the specific power dissipated by an area unit of IC microprocessors (MPUs), which reached values similar to an electrical stove burner. Because of the equivalent scaling of design, the size of the crystal in consumer and high-performance MPUs became limited, and the integration is doubling every three years instead of two (fig.1).
As it was noted previously, at a transistor level, the required power is related to the speed and voltage of the power supply Vdd. In geometric scaling, these values follow Dennard’s Law, the principle of correlated changes in the size of transistor’s elements and characteristics according to Moore’s Law. However, in reality they appeared to have reached their limits, which was particularly noted in the statement of Bob Colwell, the Director of MTO DARPA, at a conference held in March 2013 [1]. In his statement, the biggest focus was on the Vdd coming to asymptotic limit when the critical sizes are below 45 nm. Thus, further scaling is possible only with a constant voltage, so the conclusion is that Dennard’s Law is no longer valid.
According to Colwell [1], Dennard’s Law lapsed in 2005, but Moore’s Law continues working in a qualitative way: "Better, Cheaper, Faster". After 2020, Moore’s Law in its various forms must be replaced by Post-Moore’s Law of Specialization and Cleanup and this period will be characterized by a rapid growth of thermal limitations in ICs and their cost of production. After 2030, new types of electronics based on the fundamental laws of physics may emerge (fig.2). However, some leading companies are investigating new electronics based on microbiology, believing that new post-semiconductor hardware will be based on viruses and bacteria, and new operating systems will be based on DNA [2]. Nevertheless, the prospects are not yet clear even at a molecular level.
Moore’s Law was a law that drove microelectronics development, where the main driving force was steady reduction of the transistor cost in ICs, but it ceased to work when ICs reached the node size of 28 nm (fig.3) [3].
Further movement to a smaller node size (down to 7-5 nm) makes the production more expensive and the cycle of changes in node sizes becomes longer [4]. The trends of changes in node sizes were described in ITRS-2013 (fig.4) [5].
Various analytical materials mention that, at least in an individual firm, the minimal sizes change every 2 years in accordance with the roadmap of N-node.
Deviations from the price reduction trend according to Moore’s Law with increasing integration is associated with a higher cost of production. In [6], it is attributed to rising costs of masks. Whilst the cost of a 90 nm design and mask was around 1 million USD, the cost of 65 nm was 3 to 4 million USD, and these norms required 1-2 masks for each physical layer. After reaching 28 nm, the conventional method of exposing is no longer applicable. At 28 nm, one pattern formation requires three processes; at 14/16 nm, 2 pattern formations and 8 processes are required; and at 10 nm, 3 pattern formations and 21 process are required [7].
The hardships in the design and production development of new generation microelectronics encourage attempts to formulate some general trend. In particular, this was manifested during the discussion of the microelectronics and technology development by the use of a mathematical term "inflection point" as a symbol denoting deviation from the classical Moore’s Law. Depending on the area of discussion, different factors may be identified as inflection points, for example, the changes in temporal laws of microelectronics development discussed in this article. To some extent, even the discussion of the effect of Moore’s Law itself may be regarded as an inflection point.
Moore’s Law has become an encyclopedic entry. This law has been debated since the moment of its emergence, i.e. for 50 years. There are two extreme opinions. The first one is that Moore’s Law means that each technology has a certain limit of complexity (for example, depending on the number of transistors on a chip), at which the IC component cost is minimal. The second one is that Moore’s Law is only a marketing strategy for meeting the consumer demand growing at a certain pace. A faster advancement of technology is supposedly possible, but not profitable [8], in other words, it may be regarded as a tacit collusion of producers.
In fact, the strength of Moore’s Law is in its objective forecast of the technological capacity for a predictable timespan. This capacity is primarily determined by the size reduction rate of IC elements. No wonder the famous triad of microelectronics development "Smaller, Faster, Cheaper" starts with size characteristics. In the classical version of Moore’s Law, the rate of changes in the components of this triad was agreed. Currently, this agreement is disrupted, which leads to believing that Moore’s Law has lapsed. However, the qualitative characteristics of "Smaller", "Faster" and "Cheaper" as trends are still valid. Therefore, it is assumed that Moore’s Law is still effective.
Thus, the situation is assessed using the range from "it continues", through "it is partly effective" to "it has lapsed". The second assessment is supported, for example, by the author [9], who says that if Moore’s Law has ceased to be effective, it is not for technical reasons, but for economic ones. This breakdown of Moore’s Law into economic and technical parts is important. The most dubious of them is the economic part, which is regarded as the major content of Moore’s Law. This opinion was supported by the author [10] and several blogs under the same name stating that Moore’s Law ended at the node size of 28 nm. However, the viability of the 28 nm technology made him change his opinion, recognizing the effect of Moore’s Law for 28 nm, while the overall trend was characterized by the term "bifurcation" [11].
The article [12] provides some interesting statistics on the viability of Moore’s Law (unfortunately without stating the original source): 24.8% believe that Moore’s Law has ceased or will cease with the size of 7 nm; 34.7% believe that Moore’s Law continues for the production of FD SOI or FinFET based ICs; 14.4% attribute the extension of the Law to the use of 3D-structures and 12.5% to graphene; 13.6% believe that the Law will never lapse. The effect of Moore’s Law has been ardently discussed because of its importance for the informational community [13].
Moore’s Law was not some internal instruction for microelectronics industry because numerous consumers relied on it too. The termination of the Law has raised fears about further price growth of ICs in connection with the growing complexity of silicon technology. The sector’s community is trying to dispel these doubts. The key for further electronics development in the spirit of Moore’s Law will be linked with new technologies, new materials used in new devices (e.g., GaAs), modernized forms of production and scientific research.
Let us try to somewhat formalize the situation with Moore’s Law. The most important IC characteristic is the integration degree along with the certainty about its relation to the structural sizes (linear dimensions of elements, areas of cells, transistors, crystals) and certainty about its relation to functional parameters (e.g., speed) and their performance (e.g., power). In turn, one of the most important characteristics of the integration degree is its rate of increase with the certainty about the timing of changes in structural sizes and functional characteristics. Moore’s Law provided these certainties, but it has been partially disrupted or changed, so the Law partially fails.
It should be noted that the blog [14] provides an almost complete quote from Moore’s article written in the Electronics journal (1965). It says that Moore considered his law to be an extrapolation of the past production experience on a finite time interval, e.g. 10 years. In fact, Moore’s Law appeared valid substantially longer. The impact of the Law on microelectronics development is manifested in many ways. In particular, this impact is associated with a higher integration degree and node size reduction, which go on occurring.
In conclusion, we note several works, which discuss the problems associated with Moore’s Law and the general situation in the microelectronic industry. Above all, it is a series of four blogs titled "Moore’s Law is Dead?" [15–17]. In [15], it is noted that essentially Moore’s Law informs the consumers of chips on the trend in IC scaling for planning their businesses. It is noted that in its original concept, the Law (1965 and 1975) did not discuss the price, speed, power consumption or reliability. It only described a periodic increase in the number of components in a crystal associated with the increasing chip area, reduction in the size of elements and design improvements. In the heyday of microelectronics, it meant that each new generation of chips had more components, higher speed and lower cost. Currently, the number of components continues growing, but it is no longer accompanied by the diminishing size, increasing speed or reducing cost of IC components.
Moore’s Law is still working, but it no longer describes the general exponential development, therefore it raised discussions on its termination. In [16], it is noted that the rising cost of lithography causes slower transition to a new node size. Multiple lithography below 45 nm based on 193 nm sources is approaching its economic limit, though technically it does not have any insuperable obstacles.
The combination of technological and economic problems cause real limitations for the exponential growth of the microelectronics market. Perhaps, these limitations can be undone by innovations creating new type keys.
The article [17] states that Moore’s Law only highlighted a trend of continuous improvements in IC characteristics. In 1997, Moore noted that the industry driver was the low cost of the product, which made it widespread. It is possible to improve the IC functionality and scope of use with the same number of transistors. A new generation IC may be faster, or with smaller elements, or more functional, or cheaper. The general conclusion is that new generations of commercial ICs must be better. Actually, this was the objective of Moore’s Law. Moore himself admitted that he named the temporal trend of microelectronics development as a law, but its longevity was unexpected for him [18].
Scaling in microelectronics continues occurring through innovations in processes, materials and design. From 1980 to present, the transistor price has reduced by six orders of magnitude, and the market of IC applications with these transistors has grown by seven orders of magnitude [19].
The general rise in costs at all stages of IC production (fig.5) has certainly slowed down the development pace of semiconductor electronics, but it goes on with the continued scaling to 14/16 nm, 10 nm, 7 nm and 5 nm. Scaling at the level of devices is complemented by scaling at the level of systems [21], when a kind of IBM’s Virtual is being created with the scaling of IC design, production and packaging. Scaling is often associated with Moore’s Law as a major factor of microelectronics development.
Moore’s Law has become a symbol of microelectronics development. Without electronics, it would impossible to develop a modern "smart" society. Fig.6 illustrates the development stages of Moore’s Law as a reflection of electronics development.
In conclusion, we bring forward some trends in the development of contemporary microelectronics:
Continuation of 2D scaling with node sizes of 20/16/14 nm, and later – 10/7/5 nm.
Continuation of production of ICs with node sizes ≥ 28 nm.
Improvement of IC parameters according to 3P Triad (Power, Performance, Price)
Use of transistors with full depletion and competition between FinFET and FDSOI.
Development of 3D IC and 3D scaling.
Rapid development of "More than Moore" domain.
Relatively slow development of "Beyond CMOS" domain.
Focus on retooling rather than on increasing production capacity.
Elaboration of new technologies for reduction of processed wafer cost, including EUV lithography and 450 mm wafers.
Discussion of Moore’s Law in search for ways of further microelectronics development.
Use of a large number of new materials in newly developed and existing devices.
Increasing requirements of customers and end users for development of new devices, for example, wearable and consumer technologies.
Transition to a new ITRS structure – ITRS 2.0.
As it was noted previously, at a transistor level, the required power is related to the speed and voltage of the power supply Vdd. In geometric scaling, these values follow Dennard’s Law, the principle of correlated changes in the size of transistor’s elements and characteristics according to Moore’s Law. However, in reality they appeared to have reached their limits, which was particularly noted in the statement of Bob Colwell, the Director of MTO DARPA, at a conference held in March 2013 [1]. In his statement, the biggest focus was on the Vdd coming to asymptotic limit when the critical sizes are below 45 nm. Thus, further scaling is possible only with a constant voltage, so the conclusion is that Dennard’s Law is no longer valid.
According to Colwell [1], Dennard’s Law lapsed in 2005, but Moore’s Law continues working in a qualitative way: "Better, Cheaper, Faster". After 2020, Moore’s Law in its various forms must be replaced by Post-Moore’s Law of Specialization and Cleanup and this period will be characterized by a rapid growth of thermal limitations in ICs and their cost of production. After 2030, new types of electronics based on the fundamental laws of physics may emerge (fig.2). However, some leading companies are investigating new electronics based on microbiology, believing that new post-semiconductor hardware will be based on viruses and bacteria, and new operating systems will be based on DNA [2]. Nevertheless, the prospects are not yet clear even at a molecular level.
Moore’s Law was a law that drove microelectronics development, where the main driving force was steady reduction of the transistor cost in ICs, but it ceased to work when ICs reached the node size of 28 nm (fig.3) [3].
Further movement to a smaller node size (down to 7-5 nm) makes the production more expensive and the cycle of changes in node sizes becomes longer [4]. The trends of changes in node sizes were described in ITRS-2013 (fig.4) [5].
Various analytical materials mention that, at least in an individual firm, the minimal sizes change every 2 years in accordance with the roadmap of N-node.
Deviations from the price reduction trend according to Moore’s Law with increasing integration is associated with a higher cost of production. In [6], it is attributed to rising costs of masks. Whilst the cost of a 90 nm design and mask was around 1 million USD, the cost of 65 nm was 3 to 4 million USD, and these norms required 1-2 masks for each physical layer. After reaching 28 nm, the conventional method of exposing is no longer applicable. At 28 nm, one pattern formation requires three processes; at 14/16 nm, 2 pattern formations and 8 processes are required; and at 10 nm, 3 pattern formations and 21 process are required [7].
The hardships in the design and production development of new generation microelectronics encourage attempts to formulate some general trend. In particular, this was manifested during the discussion of the microelectronics and technology development by the use of a mathematical term "inflection point" as a symbol denoting deviation from the classical Moore’s Law. Depending on the area of discussion, different factors may be identified as inflection points, for example, the changes in temporal laws of microelectronics development discussed in this article. To some extent, even the discussion of the effect of Moore’s Law itself may be regarded as an inflection point.
Moore’s Law has become an encyclopedic entry. This law has been debated since the moment of its emergence, i.e. for 50 years. There are two extreme opinions. The first one is that Moore’s Law means that each technology has a certain limit of complexity (for example, depending on the number of transistors on a chip), at which the IC component cost is minimal. The second one is that Moore’s Law is only a marketing strategy for meeting the consumer demand growing at a certain pace. A faster advancement of technology is supposedly possible, but not profitable [8], in other words, it may be regarded as a tacit collusion of producers.
In fact, the strength of Moore’s Law is in its objective forecast of the technological capacity for a predictable timespan. This capacity is primarily determined by the size reduction rate of IC elements. No wonder the famous triad of microelectronics development "Smaller, Faster, Cheaper" starts with size characteristics. In the classical version of Moore’s Law, the rate of changes in the components of this triad was agreed. Currently, this agreement is disrupted, which leads to believing that Moore’s Law has lapsed. However, the qualitative characteristics of "Smaller", "Faster" and "Cheaper" as trends are still valid. Therefore, it is assumed that Moore’s Law is still effective.
Thus, the situation is assessed using the range from "it continues", through "it is partly effective" to "it has lapsed". The second assessment is supported, for example, by the author [9], who says that if Moore’s Law has ceased to be effective, it is not for technical reasons, but for economic ones. This breakdown of Moore’s Law into economic and technical parts is important. The most dubious of them is the economic part, which is regarded as the major content of Moore’s Law. This opinion was supported by the author [10] and several blogs under the same name stating that Moore’s Law ended at the node size of 28 nm. However, the viability of the 28 nm technology made him change his opinion, recognizing the effect of Moore’s Law for 28 nm, while the overall trend was characterized by the term "bifurcation" [11].
The article [12] provides some interesting statistics on the viability of Moore’s Law (unfortunately without stating the original source): 24.8% believe that Moore’s Law has ceased or will cease with the size of 7 nm; 34.7% believe that Moore’s Law continues for the production of FD SOI or FinFET based ICs; 14.4% attribute the extension of the Law to the use of 3D-structures and 12.5% to graphene; 13.6% believe that the Law will never lapse. The effect of Moore’s Law has been ardently discussed because of its importance for the informational community [13].
Moore’s Law was not some internal instruction for microelectronics industry because numerous consumers relied on it too. The termination of the Law has raised fears about further price growth of ICs in connection with the growing complexity of silicon technology. The sector’s community is trying to dispel these doubts. The key for further electronics development in the spirit of Moore’s Law will be linked with new technologies, new materials used in new devices (e.g., GaAs), modernized forms of production and scientific research.
Let us try to somewhat formalize the situation with Moore’s Law. The most important IC characteristic is the integration degree along with the certainty about its relation to the structural sizes (linear dimensions of elements, areas of cells, transistors, crystals) and certainty about its relation to functional parameters (e.g., speed) and their performance (e.g., power). In turn, one of the most important characteristics of the integration degree is its rate of increase with the certainty about the timing of changes in structural sizes and functional characteristics. Moore’s Law provided these certainties, but it has been partially disrupted or changed, so the Law partially fails.
It should be noted that the blog [14] provides an almost complete quote from Moore’s article written in the Electronics journal (1965). It says that Moore considered his law to be an extrapolation of the past production experience on a finite time interval, e.g. 10 years. In fact, Moore’s Law appeared valid substantially longer. The impact of the Law on microelectronics development is manifested in many ways. In particular, this impact is associated with a higher integration degree and node size reduction, which go on occurring.
In conclusion, we note several works, which discuss the problems associated with Moore’s Law and the general situation in the microelectronic industry. Above all, it is a series of four blogs titled "Moore’s Law is Dead?" [15–17]. In [15], it is noted that essentially Moore’s Law informs the consumers of chips on the trend in IC scaling for planning their businesses. It is noted that in its original concept, the Law (1965 and 1975) did not discuss the price, speed, power consumption or reliability. It only described a periodic increase in the number of components in a crystal associated with the increasing chip area, reduction in the size of elements and design improvements. In the heyday of microelectronics, it meant that each new generation of chips had more components, higher speed and lower cost. Currently, the number of components continues growing, but it is no longer accompanied by the diminishing size, increasing speed or reducing cost of IC components.
Moore’s Law is still working, but it no longer describes the general exponential development, therefore it raised discussions on its termination. In [16], it is noted that the rising cost of lithography causes slower transition to a new node size. Multiple lithography below 45 nm based on 193 nm sources is approaching its economic limit, though technically it does not have any insuperable obstacles.
The combination of technological and economic problems cause real limitations for the exponential growth of the microelectronics market. Perhaps, these limitations can be undone by innovations creating new type keys.
The article [17] states that Moore’s Law only highlighted a trend of continuous improvements in IC characteristics. In 1997, Moore noted that the industry driver was the low cost of the product, which made it widespread. It is possible to improve the IC functionality and scope of use with the same number of transistors. A new generation IC may be faster, or with smaller elements, or more functional, or cheaper. The general conclusion is that new generations of commercial ICs must be better. Actually, this was the objective of Moore’s Law. Moore himself admitted that he named the temporal trend of microelectronics development as a law, but its longevity was unexpected for him [18].
Scaling in microelectronics continues occurring through innovations in processes, materials and design. From 1980 to present, the transistor price has reduced by six orders of magnitude, and the market of IC applications with these transistors has grown by seven orders of magnitude [19].
The general rise in costs at all stages of IC production (fig.5) has certainly slowed down the development pace of semiconductor electronics, but it goes on with the continued scaling to 14/16 nm, 10 nm, 7 nm and 5 nm. Scaling at the level of devices is complemented by scaling at the level of systems [21], when a kind of IBM’s Virtual is being created with the scaling of IC design, production and packaging. Scaling is often associated with Moore’s Law as a major factor of microelectronics development.
Moore’s Law has become a symbol of microelectronics development. Without electronics, it would impossible to develop a modern "smart" society. Fig.6 illustrates the development stages of Moore’s Law as a reflection of electronics development.
In conclusion, we bring forward some trends in the development of contemporary microelectronics:
Continuation of 2D scaling with node sizes of 20/16/14 nm, and later – 10/7/5 nm.
Continuation of production of ICs with node sizes ≥ 28 nm.
Improvement of IC parameters according to 3P Triad (Power, Performance, Price)
Use of transistors with full depletion and competition between FinFET and FDSOI.
Development of 3D IC and 3D scaling.
Rapid development of "More than Moore" domain.
Relatively slow development of "Beyond CMOS" domain.
Focus on retooling rather than on increasing production capacity.
Elaboration of new technologies for reduction of processed wafer cost, including EUV lithography and 450 mm wafers.
Discussion of Moore’s Law in search for ways of further microelectronics development.
Use of a large number of new materials in newly developed and existing devices.
Increasing requirements of customers and end users for development of new devices, for example, wearable and consumer technologies.
Transition to a new ITRS structure – ITRS 2.0.
Readers feedback
