Hardware technology discussion replacements in the 3rd edition:
1.
3rd edition: At this point, the speaker described numerous problems with the Moore’s Law extension, and the new approach suggested “by our research group”. In their MolBars, a usual multilayer CMOS circuit would be complemented with nanowire crossbar [1] with a molecular nanodevice self-assembled at each crosspoint.
2nd edition: “Many believe that the power dissipation in microprocessor chips will become too large for the resulting heat to be manageable. However, the ongoing transfer of the FETs from their traditional ‘planar’ variety to a new, ‘surrounding-gate’ version should reduce the dissipation very significantly. Moreover, my microprocessor-architecture friends are telling me there are very significant reserves left for power reduction in their court as well – even though they would prefer to avoid the hard work mining them.
“In my humble view, a much more difficult hurdle for Moore’s Law to overcome is the integrated circuit patterning, using photolithography. Historically, the circuit scaling was achieved mostly by reducing the wavelength of light used for the lithography; however, this process had bottomed out at 193 nm (nanometers), because there is no good candidate to replace the argon-fluorine excimer lasers used as sources of this ultraviolet ‘light’. There are some tricks we still can do, even at the currently used wavelength, but they will hardly enable us to reduce the most important dimensions, such as the transistor channel length, below 30 nm or so with the necessary accuracy. I believe there is a radical way to change the situation, using a combination of new devices – molecular ‘latching switches’ – and a new circuit topology. Let me start with the devices.”
2.
3rd edition: …with the functionality necessary for Molbar operation.
2nd edition: …with a very special property. When stretched between two metal electrodes, these molecules could be switched, by applying a certain voltage between the electrodes, between two different configurations: one with a relatively high, and the second with a very low electric conductance. If no voltage was applied, the molecule would maintain its current state – hence the term “latching switch”.
3.
3rd edition: Here Kravtsov started to describe in detail how his MolBars would improve the density and performance of digital memory and logic circuits, but then quickened his pace – he had only ten minutes left of his allotted time.
2nd edition: The simplest, and most efficient structure to engage them is the ‘crossbar’ – the simplest rectangular matrix structure.” He pulled up the next slide depicting two layers of parallel nanowires, placed on top of each other at a right angle. “Now, imagine we have placed one molecular latching switch on each crosspoint so that it connects one wire of the top layer with one wire of the lower layer. Such crossbars could be, first of all, the basis of ultra-dense memories, because we could change the device’s conducting state, and then read it out electronically by applying relatively low voltages to the proper wires of both layers and measuring the resulting current.”
At this point, Kira was ready to protest. The whole concept made sense only if the crossbar was very dense, with wires only a few nanometers wide. But how could such wires be formed? And how could they be connected to the usual CMOS-circuit wires, with widths of at least a hundred nanometers? As if the speaker had heard Kira’s internal screams, he pulled up a slide showing a possible area-distributed interface between a nanowire crossbar and an underlying CMOS circuit, with a square array of sharp vertical pins in its top layer, with a small angle between them. The resulting large-scale periodicity (Kravtsov said such moiré patterns had been discovered in the textile industry long ago) would enable the CMOS circuit to contact each and every nanowire of each crossbar layer, and hence each molecular nanodevice at their crosspoints. He called this hybrid CMOS/nanowire-crossbar/molecular-device circuit a “Molecular Crossbar,” or MolBar for short.
The first of Kira’s questions was apparently also obvious because at this moment one of the people in the front row asked it: “Sorry for interrupting, but how would you fabricate these nanowires if, as you said, patterning technologies are going to bottom out at about 30 nm?”
“That statement referred only to the usual lithographic techniques, which allow us to define virtually arbitrary patterns,” Kravtsov responded. “However, for a crossbar, we need only sets of parallel nanowires, and for such a simple pattern, there are several special methods which are already approaching a 10-nm resolution.”
Kravtsov spent almost five minutes describing these patterning techniques, but Kira could not wait for the discussion of possible applications of this idea. “The most obvious application of MolBars would be ultra-dense ‘resistive’ memories. If our predictions of crossbar scaling are correct, the density of such memory chips will be unprecedented: above one Terabit (1012 bits) per square centimeter.”
After discussing a specific software trick, doubling the address space necessary for a unique nanodevice selection in such memories, he continued, “A less evident application opportunity is in digital logic circuits, for example in FPGAs.”
“LL, what are the FPGAs?” asked somebody from a front row in unaccented American English.
“Oh, sorry Bill, you are a physicist…”
“As quite a few people in this room are,” Bill snapped.
“OK, sorry again. FPGA is the acronym of Field-Programmable Gate Array. This is a general-purpose silicon chip with a mass of logic gates and memory cells, and with the means to configure them for a specific purpose already after the fabrication. This is a very price-effective approach, though the resulting circuits usually have lower performance than integrated circuits with a more dedicated structure, such as the CPU [2] in your PC.” Here he glanced at Bill, who nodded to indicate he at least knew what the CPU was. “In MolBar-based FPGAs, the configuration would be done using the same compact crosspoint devices, again enabling an extremely high density and, due to the corresponding reduction of parasitic capacitances, a very high speed.
4.
3rd edition: “Well, sort of, but here is what I do not know. Let us imagine that this molecular self-assembly of yours does not, in fact, work.”
2nd edition: “Well, sort of, but here is what I do not know. I have heard from one of my professors that the current difficulties of circuit patterning, which were outlined in your talk, will soon be solved by a new type of lithography using ‘extreme ultraviolet’ radiation. Is there any truth in this advertising?”
“Briefly speaking, not much,” LL said. “That ‘EUV lithography’ uses 13-nm radiation from plasma sources with power way too low for practical applications. There is also a big problem of refractive optics contamination and several other serious challenges. I believe the obstacles on the EUVL’s path to commercial production lines are so large that even if and when they have been overcome, optimistically in fifteen years or so, the resulting patterning systems will be too expensive to be economically viable.”
“I see. Next, us imagine that this molecular self-assembly of yours does not, in fact, work.”
5.
3rd edition: “Well,” Kravtsov said somewhat reluctantly, “actually, there are such devices, perhaps even too many kinds of them – mostly thin amorphous layers of various metal oxides, sandwiched between metal electrodes.”
2nd edition: “Well,” Kravtsov said somewhat reluctantly, “actually, there are such devices, perhaps even too many kinds of them – mostly thin amorphous layers of various metal oxides, sandwiched between metal electrodes. There are even many names for them: ‘latching switches’, ‘programmable diodes’, ‘switched resistors’, ‘memistors’, and on and on, because they were discovered and re-discovered again and again, starting at least from the 1950s.”
6.
3rd edition: My problem with all these devices is that their operation is based on shifts of just a few ions inside the oxide.
2nd edition: My problem with all these devices is that the voltage-induced switching of their electric conductance is based on shifts of just a few ions inside the oxide, either creating or suppressing a highly conductive path between the electrodes.
7.
3rd edition: And why can’t you use, in the meantime, floating-gate cells, such as those in flash memories, for this purpose?
2nd edition: And why can’t you use, in the meantime, floating-gate cells, such as those in flash memories,[3] for this purpose?
8.
3rd edition: “But back to your question: indeed, floating-gate memory cells may be used in circuits similar to our MolBar. My main problem…”
2nd edition: “Now to your question: indeed, floating-gate memory cells may be used in circuits similar to our MolBar. As far as I know, the first paper on this was published in 1995 by a Caltech group led by that famous guy Carver Mead.”
“So, this CMOS-to-crossbar interface of yours (an idea that I just love!) had been done before? I don’t believe you mentioned that in your talk.”
“No, for small crossbars (and I believe the original Caltech crossbar had just four crosspoints) you don’t need our interface: you may simply widen each narrow wire of the crossbar gradually until its width fits that of the external wiring. In my talk, I was discussing large crossbars, which are needed for serious applications. For these, such a peripheral interface would not work, and you would need to use an area-distributed one like ours.
“But back to the floating-gate transistors. My main problem…”
[1] In this context, crossbar is a set of two adjacent layers of parallel wires, with the wires of one layer running across those of the counterpart layer.
[2] The acronym for Central Processing Unit – the computer’s main information-processing chip, sometimes called the microprocessor.
[3] Each cell of the ubiquitous flash memories may have two (and sometimes more) states, which differ by the electric charge of a special electrode, called the floating gate, and may be switched by applying sufficiently large voltages that change this charge.
3rd edition: At this point, the speaker described numerous problems with the Moore’s Law extension, and the new approach suggested “by our research group”. In their MolBars, a usual multilayer CMOS circuit would be complemented with nanowire crossbar [1] with a molecular nanodevice self-assembled at each crosspoint.
2nd edition: “Many believe that the power dissipation in microprocessor chips will become too large for the resulting heat to be manageable. However, the ongoing transfer of the FETs from their traditional ‘planar’ variety to a new, ‘surrounding-gate’ version should reduce the dissipation very significantly. Moreover, my microprocessor-architecture friends are telling me there are very significant reserves left for power reduction in their court as well – even though they would prefer to avoid the hard work mining them.
“In my humble view, a much more difficult hurdle for Moore’s Law to overcome is the integrated circuit patterning, using photolithography. Historically, the circuit scaling was achieved mostly by reducing the wavelength of light used for the lithography; however, this process had bottomed out at 193 nm (nanometers), because there is no good candidate to replace the argon-fluorine excimer lasers used as sources of this ultraviolet ‘light’. There are some tricks we still can do, even at the currently used wavelength, but they will hardly enable us to reduce the most important dimensions, such as the transistor channel length, below 30 nm or so with the necessary accuracy. I believe there is a radical way to change the situation, using a combination of new devices – molecular ‘latching switches’ – and a new circuit topology. Let me start with the devices.”
2.
3rd edition: …with the functionality necessary for Molbar operation.
2nd edition: …with a very special property. When stretched between two metal electrodes, these molecules could be switched, by applying a certain voltage between the electrodes, between two different configurations: one with a relatively high, and the second with a very low electric conductance. If no voltage was applied, the molecule would maintain its current state – hence the term “latching switch”.
3.
3rd edition: Here Kravtsov started to describe in detail how his MolBars would improve the density and performance of digital memory and logic circuits, but then quickened his pace – he had only ten minutes left of his allotted time.
2nd edition: The simplest, and most efficient structure to engage them is the ‘crossbar’ – the simplest rectangular matrix structure.” He pulled up the next slide depicting two layers of parallel nanowires, placed on top of each other at a right angle. “Now, imagine we have placed one molecular latching switch on each crosspoint so that it connects one wire of the top layer with one wire of the lower layer. Such crossbars could be, first of all, the basis of ultra-dense memories, because we could change the device’s conducting state, and then read it out electronically by applying relatively low voltages to the proper wires of both layers and measuring the resulting current.”
At this point, Kira was ready to protest. The whole concept made sense only if the crossbar was very dense, with wires only a few nanometers wide. But how could such wires be formed? And how could they be connected to the usual CMOS-circuit wires, with widths of at least a hundred nanometers? As if the speaker had heard Kira’s internal screams, he pulled up a slide showing a possible area-distributed interface between a nanowire crossbar and an underlying CMOS circuit, with a square array of sharp vertical pins in its top layer, with a small angle between them. The resulting large-scale periodicity (Kravtsov said such moiré patterns had been discovered in the textile industry long ago) would enable the CMOS circuit to contact each and every nanowire of each crossbar layer, and hence each molecular nanodevice at their crosspoints. He called this hybrid CMOS/nanowire-crossbar/molecular-device circuit a “Molecular Crossbar,” or MolBar for short.
The first of Kira’s questions was apparently also obvious because at this moment one of the people in the front row asked it: “Sorry for interrupting, but how would you fabricate these nanowires if, as you said, patterning technologies are going to bottom out at about 30 nm?”
“That statement referred only to the usual lithographic techniques, which allow us to define virtually arbitrary patterns,” Kravtsov responded. “However, for a crossbar, we need only sets of parallel nanowires, and for such a simple pattern, there are several special methods which are already approaching a 10-nm resolution.”
Kravtsov spent almost five minutes describing these patterning techniques, but Kira could not wait for the discussion of possible applications of this idea. “The most obvious application of MolBars would be ultra-dense ‘resistive’ memories. If our predictions of crossbar scaling are correct, the density of such memory chips will be unprecedented: above one Terabit (1012 bits) per square centimeter.”
After discussing a specific software trick, doubling the address space necessary for a unique nanodevice selection in such memories, he continued, “A less evident application opportunity is in digital logic circuits, for example in FPGAs.”
“LL, what are the FPGAs?” asked somebody from a front row in unaccented American English.
“Oh, sorry Bill, you are a physicist…”
“As quite a few people in this room are,” Bill snapped.
“OK, sorry again. FPGA is the acronym of Field-Programmable Gate Array. This is a general-purpose silicon chip with a mass of logic gates and memory cells, and with the means to configure them for a specific purpose already after the fabrication. This is a very price-effective approach, though the resulting circuits usually have lower performance than integrated circuits with a more dedicated structure, such as the CPU [2] in your PC.” Here he glanced at Bill, who nodded to indicate he at least knew what the CPU was. “In MolBar-based FPGAs, the configuration would be done using the same compact crosspoint devices, again enabling an extremely high density and, due to the corresponding reduction of parasitic capacitances, a very high speed.
4.
3rd edition: “Well, sort of, but here is what I do not know. Let us imagine that this molecular self-assembly of yours does not, in fact, work.”
2nd edition: “Well, sort of, but here is what I do not know. I have heard from one of my professors that the current difficulties of circuit patterning, which were outlined in your talk, will soon be solved by a new type of lithography using ‘extreme ultraviolet’ radiation. Is there any truth in this advertising?”
“Briefly speaking, not much,” LL said. “That ‘EUV lithography’ uses 13-nm radiation from plasma sources with power way too low for practical applications. There is also a big problem of refractive optics contamination and several other serious challenges. I believe the obstacles on the EUVL’s path to commercial production lines are so large that even if and when they have been overcome, optimistically in fifteen years or so, the resulting patterning systems will be too expensive to be economically viable.”
“I see. Next, us imagine that this molecular self-assembly of yours does not, in fact, work.”
5.
3rd edition: “Well,” Kravtsov said somewhat reluctantly, “actually, there are such devices, perhaps even too many kinds of them – mostly thin amorphous layers of various metal oxides, sandwiched between metal electrodes.”
2nd edition: “Well,” Kravtsov said somewhat reluctantly, “actually, there are such devices, perhaps even too many kinds of them – mostly thin amorphous layers of various metal oxides, sandwiched between metal electrodes. There are even many names for them: ‘latching switches’, ‘programmable diodes’, ‘switched resistors’, ‘memistors’, and on and on, because they were discovered and re-discovered again and again, starting at least from the 1950s.”
6.
3rd edition: My problem with all these devices is that their operation is based on shifts of just a few ions inside the oxide.
2nd edition: My problem with all these devices is that the voltage-induced switching of their electric conductance is based on shifts of just a few ions inside the oxide, either creating or suppressing a highly conductive path between the electrodes.
7.
3rd edition: And why can’t you use, in the meantime, floating-gate cells, such as those in flash memories, for this purpose?
2nd edition: And why can’t you use, in the meantime, floating-gate cells, such as those in flash memories,[3] for this purpose?
8.
3rd edition: “But back to your question: indeed, floating-gate memory cells may be used in circuits similar to our MolBar. My main problem…”
2nd edition: “Now to your question: indeed, floating-gate memory cells may be used in circuits similar to our MolBar. As far as I know, the first paper on this was published in 1995 by a Caltech group led by that famous guy Carver Mead.”
“So, this CMOS-to-crossbar interface of yours (an idea that I just love!) had been done before? I don’t believe you mentioned that in your talk.”
“No, for small crossbars (and I believe the original Caltech crossbar had just four crosspoints) you don’t need our interface: you may simply widen each narrow wire of the crossbar gradually until its width fits that of the external wiring. In my talk, I was discussing large crossbars, which are needed for serious applications. For these, such a peripheral interface would not work, and you would need to use an area-distributed one like ours.
“But back to the floating-gate transistors. My main problem…”
[1] In this context, crossbar is a set of two adjacent layers of parallel wires, with the wires of one layer running across those of the counterpart layer.
[2] The acronym for Central Processing Unit – the computer’s main information-processing chip, sometimes called the microprocessor.
[3] Each cell of the ubiquitous flash memories may have two (and sometimes more) states, which differ by the electric charge of a special electrode, called the floating gate, and may be switched by applying sufficiently large voltages that change this charge.