Periodic Reporting for period 3 - SPICE (Spintronic-Photonic Integrated Circuit platform for novel Electronics)
Okres sprawozdawczy: 2019-04-01 do 2020-12-31
Exponential growth of data means that it is essential to decrease the energy consumption of data storage. Currently, various memory technologies are used, including, hard disks, flash and random-access-memory. Some technologies are non-volatile, which means they do not consume power once the data is stored, but these typically consume a lot of power during writing and are slow. Many fast technologies are often volatile.
It would be ideal to have one type of memory, one that is fast and cheap, low-power, and non-volatile. Spintronic memory is based on magnetics, and can potentially combine this.
A promising trend is the discovery of switching the magnetization in thin magnetic films by short laser pulses, at very low energy. This is a new way to write data into spintronic memory using light, and promises terabit-per-second bandwidth, at unprecedented low energy consumption. The open question is how to realize such an approach on an integrated circuit.
SPICE proposes a novel technology platform that integrates photonics and spintronics. Firstly, new magneto-optic materials were developed, that can switch under illumination. These were then integrated into spintronic memory elements. Secondly, energy-efficient optical networks on a chip were realized, to write data into these elements. The overall objective was to improve the speed and energy consumption of current memory by two orders of magnitude.
The conclusion of the SPICE action is that such magneto-optic materials have been realized, which can switch with a single optical pulse. Using these materials, magnetic tunnel junction stacks have been realized and further processed into electrically contacted pillars, with a top contact made of a transparent electrode, so the pillars can still be illuminated optically. Also the elements for a photonic switch networks have been realized in this action, most notably energy-efficient switches and elements to couple the light from the optical layer into the memory element. We theoretically assessed that such an approach would be competitive with existing memory technologies, in terms of energy efficiency.
It would be ideal to have one type of memory, one that is fast and cheap, low-power, and non-volatile. Spintronic memory is based on magnetics, and can potentially combine this.
A promising trend is the discovery of switching the magnetization in thin magnetic films by short laser pulses, at very low energy. This is a new way to write data into spintronic memory using light, and promises terabit-per-second bandwidth, at unprecedented low energy consumption. The open question is how to realize such an approach on an integrated circuit.
SPICE proposes a novel technology platform that integrates photonics and spintronics. Firstly, new magneto-optic materials were developed, that can switch under illumination. These were then integrated into spintronic memory elements. Secondly, energy-efficient optical networks on a chip were realized, to write data into these elements. The overall objective was to improve the speed and energy consumption of current memory by two orders of magnitude.
The conclusion of the SPICE action is that such magneto-optic materials have been realized, which can switch with a single optical pulse. Using these materials, magnetic tunnel junction stacks have been realized and further processed into electrically contacted pillars, with a top contact made of a transparent electrode, so the pillars can still be illuminated optically. Also the elements for a photonic switch networks have been realized in this action, most notably energy-efficient switches and elements to couple the light from the optical layer into the memory element. We theoretically assessed that such an approach would be competitive with existing memory technologies, in terms of energy efficiency.
Over the duration of this project, the foundations of a fully integrated photonic-spintronic memory technology have been laid out:
The first area is the development of magnetic materials that can be optically switched. Work has converged on using a material stack based on an optically-switchable Tb/Co multilayer, and a conventional ferromagnetic electrode FeCoB layer. We have shown that this layer can be switched by single picosecond optical pulses, which is a key achievement for the proposed technology. The energy density needed for switching is comparable to best-in-class results of similar technologies. The technology has been evaluated including necessary thermal anneal steps, which is required for further integration.
The second area is the fabrication of the memory elements. A process flow was first established and validated. Secondly, indium tin oxide was selected for the top contact, to allow for optical access, and a process to fabricate magnetic tunnel junction (MTJ) elements with optically transparent top contacts was established. MTJ elements were realized that could be switched optically, by a single short pulse, and read out electrically, by a change in resistance, thereby confirming usefulness as a memory element. The process is available for third parties.
The third area is the development of the photonic distribution layer. An energy-efficient switch network for the optical pulses was developed, based on optimized silicon pin modulators. Gated operation, to reduce the energy consumption, was experimentally achieved, and it was calculated that such networks could work at 100 fJ-per-bit levels. Elements to couple the optical pulse from the distribution layer into the memory element were designed and realized, with close to diffraction-limited focal spots. Active control of the polarization was achieved, for polarization based switching, and solutions for deterministic toggle-based switching were designed and implemented in the circuits. Initial work on reducing the energy required for switching, using plasmonic concentrators, was started. Switches are available for third parties.
The fourth area is the development of design tools. A material simulation software tool has been further optimized for spintronic applications. The tool has been applied to calculating, o.a. the temperature dependence of the tunnel magnetoresistance. A simulation flow, including this tool, and other, open-access, tools, was established, to include magnetization dynamics. These results can then be exported to circuit-level electronic-design-automation tools. This latter work resulted in a compact model, that was shared with the community for open use.
These results converged into our Demonstrator effort, where a chip with an MTJ array was coupled to a photonic switching network. Due to limitations in the power handling capacity of the silicon photonic chip, the optical power that could be delivered was not enough to achieve switching in a fully integrated configuration. The operation of the individual chips was validated, though, and the path towards full integration was outlined.
The first area is the development of magnetic materials that can be optically switched. Work has converged on using a material stack based on an optically-switchable Tb/Co multilayer, and a conventional ferromagnetic electrode FeCoB layer. We have shown that this layer can be switched by single picosecond optical pulses, which is a key achievement for the proposed technology. The energy density needed for switching is comparable to best-in-class results of similar technologies. The technology has been evaluated including necessary thermal anneal steps, which is required for further integration.
The second area is the fabrication of the memory elements. A process flow was first established and validated. Secondly, indium tin oxide was selected for the top contact, to allow for optical access, and a process to fabricate magnetic tunnel junction (MTJ) elements with optically transparent top contacts was established. MTJ elements were realized that could be switched optically, by a single short pulse, and read out electrically, by a change in resistance, thereby confirming usefulness as a memory element. The process is available for third parties.
The third area is the development of the photonic distribution layer. An energy-efficient switch network for the optical pulses was developed, based on optimized silicon pin modulators. Gated operation, to reduce the energy consumption, was experimentally achieved, and it was calculated that such networks could work at 100 fJ-per-bit levels. Elements to couple the optical pulse from the distribution layer into the memory element were designed and realized, with close to diffraction-limited focal spots. Active control of the polarization was achieved, for polarization based switching, and solutions for deterministic toggle-based switching were designed and implemented in the circuits. Initial work on reducing the energy required for switching, using plasmonic concentrators, was started. Switches are available for third parties.
The fourth area is the development of design tools. A material simulation software tool has been further optimized for spintronic applications. The tool has been applied to calculating, o.a. the temperature dependence of the tunnel magnetoresistance. A simulation flow, including this tool, and other, open-access, tools, was established, to include magnetization dynamics. These results can then be exported to circuit-level electronic-design-automation tools. This latter work resulted in a compact model, that was shared with the community for open use.
These results converged into our Demonstrator effort, where a chip with an MTJ array was coupled to a photonic switching network. Due to limitations in the power handling capacity of the silicon photonic chip, the optical power that could be delivered was not enough to achieve switching in a fully integrated configuration. The operation of the individual chips was validated, though, and the path towards full integration was outlined.
SPICE delivered some interesting and potentially high-impact results.
• We have achieved optical switching by a single picosecond optical pulse Tb/Co multilayers at energy densities comparable with state of the art in similar technologies. We have succeeded in integrating these layers into an MTJ layerstack. This implies that for the first time we have realized a practical material platform for optically-controlled MTJs.
• We have designed and realized a new process, which can add optically transparent top contacts to an MTJ. This was implemented on MTJs containing Tb/Co multilayers, and we have shown that these can be switched by a single optical pulse, while being read electrically. This means that we have established the full process for optically-switchable, and electrically-readable MTJ elements.
• We have realized energy-efficient switch networks and grating couplers in a silicon photonics platform. For the switching of short optical pulses, these switches were shown to be superior as compared to state of the art. The impact is that we now have a CMOS-compatible way of controlling the optical pulses, at low energy consumption, and potentially high bitrates.
• For the simulation effort, we have set up a flow of commercially-available and open-source tools, to be able to simulate MTJs from the physics to the circuit level. The compatibility with photonic design tools implies a full spintronic-photonic-electronic co-simulation environment.
• A theoretical assessment has shown that, based on existing and available technologies, the SPICE memory technology can be competitive with selected existing memory technologies, in terms of bandwidth and power consumption, albeit not in footprint. The impact of such a result is that, for the first time, the feasibility of optically-switched MTJ-based memory technology has been shown.
We thus have realized a technology concept that has a clear path out of the lab, due to the compatibility of the used processes with mature technologies, and that has a clear motivation to go out of the lab, due to the competitiveness with respect to applications. Challenges, however, have also been identified, but these do not seem unsurmountable.
• We have achieved optical switching by a single picosecond optical pulse Tb/Co multilayers at energy densities comparable with state of the art in similar technologies. We have succeeded in integrating these layers into an MTJ layerstack. This implies that for the first time we have realized a practical material platform for optically-controlled MTJs.
• We have designed and realized a new process, which can add optically transparent top contacts to an MTJ. This was implemented on MTJs containing Tb/Co multilayers, and we have shown that these can be switched by a single optical pulse, while being read electrically. This means that we have established the full process for optically-switchable, and electrically-readable MTJ elements.
• We have realized energy-efficient switch networks and grating couplers in a silicon photonics platform. For the switching of short optical pulses, these switches were shown to be superior as compared to state of the art. The impact is that we now have a CMOS-compatible way of controlling the optical pulses, at low energy consumption, and potentially high bitrates.
• For the simulation effort, we have set up a flow of commercially-available and open-source tools, to be able to simulate MTJs from the physics to the circuit level. The compatibility with photonic design tools implies a full spintronic-photonic-electronic co-simulation environment.
• A theoretical assessment has shown that, based on existing and available technologies, the SPICE memory technology can be competitive with selected existing memory technologies, in terms of bandwidth and power consumption, albeit not in footprint. The impact of such a result is that, for the first time, the feasibility of optically-switched MTJ-based memory technology has been shown.
We thus have realized a technology concept that has a clear path out of the lab, due to the compatibility of the used processes with mature technologies, and that has a clear motivation to go out of the lab, due to the competitiveness with respect to applications. Challenges, however, have also been identified, but these do not seem unsurmountable.