Atomic layer deposition (ALD) used to be considered too slow to be of practical use in semiconductor manufacturing, but it has emerged as a critical tool for both transistor and interconnect fabrication at the most advanced nodes.

ALD can be speeded up somewhat, but the real shift is the rising value of precise composition and thickness control at the most advanced nodes, which makes the extra time spent on deposition worthwhile.

ALD is a close cousin of chemical vapor deposition, initially introduced in high volume to the semiconductor industry for hafnium oxide (high-k) gate dielectrics. Both CVD and ALD are inherently conformal processes. Deposition occurs on all surfaces exposed to a precursor gas. In ALD, though, the reaction is self-limiting.

The process works like this: First, a precursor gas (A) is introduced into the process chamber, where it adsorbs onto all available substrate sites. No further adsorption occurs once all surface sites are occupied. An inert purge gas, typically nitrogen or argon, flushes out any remaining precursor gas, then a second precursor (B) is introduced. Precursor B reacts with the chemisorbed precursor A to produce the desired film. Once all of the adsorbed molecules are consumed, the reaction stops. After a second purge step, the cycle repeats.

ALD opportunities expand as features shrink
The step-by-step nature of ALD is both its strength and its weakness. Depositing one monolayer at a time gives manufacturers extremely precise thickness control. Using different precursor gases in different ratios can tune the film composition. Unfortunately, the repeated precursor/purge gas cycles take a lot of time. In an interview, CEA-Leti researcher Rémy Gassilloud estimated that in a single wafer process, two minutes per wafer is the maximum cost-effective process time. But two minutes is only enough time to deposit about a 2nm-thick film.

Some process adjustments can improve throughput. Silicon dioxide ALD often uses large furnaces to process many wafers at once. Plasma activation can ionize reagents and accelerate film formation. Still, Gassilloud estimates that 10nm is the maximum practical thickness for ALD films.

As transistors shrink, though, the number of layers in that thickness range is increasing. Transistor structures also are becoming more complex, requiring deposition on vertical surfaces, into deep trenches, and other places not readily accessible by line-of-sight PVD methods. Replacement gates for gate-all-around transistors, for instance, need a process that can fill nanometer-scale cavities.

As noted above, HfO₂ was the first successful application of ALD in semiconductor manufacturing. Its precursors, HfCl₄ and water, are both chemically simple small molecules, whose by-products are volatile and easily removed. Such simple chemistries are the exception, though. ALD of silicon dioxide typically uses aminosilane precursors.⁠[1] Metal nitrides often have complex metal-organic precursor gases. Gassilloud noted that ligands might be added to a precursor molecule to change its vapor pressure or reactivity, or to facilitate adhesion to the substrate. In selective deposition processes, discussed below, ligands might improve selectivity between growth and non-growth surfaces. These larger molecules can be difficult to insinuate into smaller features, and byproducts can be difficult to remove. Complex byproducts can also become a contamination source.

One of the advantages of ALD is its very low process temperature, typically between 200°C and 300°C. It is thermally compatible with both transistor and interconnect processes in CMOS, as well as with deposition on plastic and other novel substrates. Even so, Aditya Kumar and colleagues at GlobalFoundries showed that precise temperature control is important.^⁠[2] TDMAT (tetrakis- dimethylamino titanium) condensation in a TiN deposition process was a significant source of particle defects. To maintain the desired process temperature, both the precursor and purge gas temperatures matter. Introducing cold purge gas into a warm process chamber can cause rapid condensation.

As ALD has become a mainstream process, the industry has found applications for it beyond core device materials, in a variety of sacrificial and spacer layers. For example, double- and quadruple-patterning schemes often use ALD for “pitch-doubling.” By depositing a spacer material on either side of a patterned “mandrel,” then removing the mandrel, the process can cut the original pitch in half without the need for an additional, more costly lithography step.^⁠[3]

Fig. 1: Self-aligned double patterning with ALD spacers. Source: Creative Commons

Depositing a doped oxide on the vertical silicon fins of a finFET device is a less directional and less damaging alternative to ion implantation.[^⁠4]

Selective deposition brings lateral control
These last two examples depend on surface characteristics to mediate deposition. A precursor might adhere more readily to a hard mask than to the underlying material. The vertical face of a silicon fin might offer more (or fewer) adsorption sites than the horizontal face. Selective deposition on more complicated structures may require a pre-deposited growth template, functionalizing substrate regions to encourage or discourage growth. Selective deposition is especially important in interconnect applications. In general, though, a comprehensive review by Rong Chen and colleagues at Huazhong University of Science and Technology explained that selective deposition methods need to replenish the template material as the film grows while needing a mechanism to selectively remove the unwanted material.⁠[5]

For example, tungsten preferentially deposits on silicon relative to SiO₂, but the selectivity diminishes after only a few cycles. Researchers at North Carolina State University successfully re-passivated the oxide by incorporating hydrogen into the tungsten precursor.[⁠6] Similarly, a group at Eindhoven University of Technology found that SiO₂ preferentially deposited on SiO₂ relative to other oxides for only 10 to 15 cycles. A so-called ABC-cycle — adding acetylacetone (“Inhibitor A”) as an inhibitor every 5 to 10 cycles — restored selectivity.⁠[7]

Alternatively, or in addition, atomic layer etching (ALE) might be used to remove unwanted material. ALE operates in the same step-by-step manner as ALD. The first half of a cycle reacts with the existing surface, weakening the bond to the underlying material. Then, a second step — typically ion bombardment — removes the weakened layer. For example, in ALE etching of silicon, chlorine gas reacts with the surface to form various SiCl_x compounds. The chlorination process weakens the inter-silicon bonds between the surface and the bulk, and the chlorinated layer is easily sputtered away. The layer-by-layer nature of ALE depends on preferential removal of the surface material relative to the bulk (SiCl_x vs. Si in this case). The “ALE window” is the combination of energy and temperature at which the surface layer is completely removed without damaging the underlying material.

Somewhat counter-intuitively, Keren Kanarik and colleagues at Lam Research found that higher ion energies actually expanded the ALE window for silicon etching. High ion energies with short exposure times delayed the onset of silicon sputtering relative to conventional RIE.[8]

Adding and subtracting, one atomic layer at a time
For a long time, the semiconductor industry has been looking for alternatives to process schemes that deposit material, pattern it, then etch most of it away. Wouldn’t it be simpler to only deposit the material we will ultimately need? Meanwhile, atomic layer deposition has been filling the spaces under nanosheets and inside cavities. Bulk deposition and etch tools are still with us, and will be for the foreseeable future. In more and more cases, though, those tools provide the frame while ALD and ALE processes fill in the details.

Correction: Corrected attribution of the work on ABC cycles and selective deposition of SiO₂.

References

1. Wenling Li, et al., “Impact of aminosilane and silanol precursor structure on atomic layer deposition process,”Applied Surface Science, Vol 621, 2023,156869, https://doi.org/10.1016/j.apsusc.2023.156869.
2. Kumar, et al., “ALD TiN Surface Defect Reduction for 12nm and Beyond Technologies,” 2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC), Saratoga Springs, NY, USA, 2020, pp. 1-4, doi: 10.1109/ASMC49169.2020.9185271.
3. Shohei Yamauchi, et al., “Extendibility of self-aligned type multiple patterning for further scaling”, Proc. SPIE 8682, Advances in Resist Materials and Processing Technology XXX, 86821D (29 March 2013); https://doi.org/10.1117/12.2011953
4. Kalkofen, et al., “Atomic layer deposition of phosphorus oxide films as solid sources for doping of semiconductor structures,” 2018 IEEE 18th International Conference on Nanotechnology (IEEE-NANO), Cork, Ireland, 2018, pp. 1-4, doi: 10.1109/NANO.2018.8626235.
5. Rong Chen et al., “Atomic level deposition to extend Moore’s law and beyond,” 2020 Int. J. Extrem. Manuf. 2 022002 DOI 10.1088/2631-7990/ab83e0
6. B Kalanyan, et al., “Using hydrogen to expand the inherent substrate selectivity window during tungsten atomic layer deposition,” 2016 Chem. Mater. 28 117–26 https://doi.org/10.1021/acs.chemmater.5b03319
7. Alfredo Mameli et al., “Area-Selective Atomic Layer Deposition of SiO2 Using Acetylacetone as a Chemoselective Inhibitor in an ABC-Type Cycle” ACS Nano 2017, 11, 9, 9303–9311. https://doi.org/10.1021/acsnano.7b04701
8. Keren J. Kanarik, et al., “Universal scaling relationship for atomic layer etching,” J. Vac. Sci. Technol. A 39, 010401 (2021); doi: 10.1116/6.0000762

The post Enabling Advanced Devices With Atomic Layer Processes appeared first on Semiconductor Engineering.

Atomic layer deposition (ALD) used to be considered too slow to be of practical use in semiconductor manufacturing, but it has emerged as a critical tool for both transistor and interconnect fabrication at the most advanced nodes.

ALD can be speeded up somewhat, but the real shift is the rising value of precise composition and thickness control at the most advanced nodes, which makes the extra time spent on deposition worthwhile.

ALD is a close cousin of chemical vapor deposition, initially introduced in high volume to the semiconductor industry for hafnium oxide (high-k) gate dielectrics. Both CVD and ALD are inherently conformal processes. Deposition occurs on all surfaces exposed to a precursor gas. In ALD, though, the reaction is self-limiting.

The process works like this: First, a precursor gas (A) is introduced into the process chamber, where it adsorbs onto all available substrate sites. No further adsorption occurs once all surface sites are occupied. An inert purge gas, typically nitrogen or argon, flushes out any remaining precursor gas, then a second precursor (B) is introduced. Precursor B reacts with the chemisorbed precursor A to produce the desired film. Once all of the adsorbed molecules are consumed, the reaction stops. After a second purge step, the cycle repeats.

ALD opportunities expand as features shrink
The step-by-step nature of ALD is both its strength and its weakness. Depositing one monolayer at a time gives manufacturers extremely precise thickness control. Using different precursor gases in different ratios can tune the film composition. Unfortunately, the repeated precursor/purge gas cycles take a lot of time. In an interview, CEA-Leti researcher Rémy Gassilloud estimated that in a single wafer process, two minutes per wafer is the maximum cost-effective process time. But two minutes is only enough time to deposit about a 2nm-thick film.

Some process adjustments can improve throughput. Silicon dioxide ALD often uses large furnaces to process many wafers at once. Plasma activation can ionize reagents and accelerate film formation. Still, Gassilloud estimates that 10nm is the maximum practical thickness for ALD films.

As transistors shrink, though, the number of layers in that thickness range is increasing. Transistor structures also are becoming more complex, requiring deposition on vertical surfaces, into deep trenches, and other places not readily accessible by line-of-sight PVD methods. Replacement gates for gate-all-around transistors, for instance, need a process that can fill nanometer-scale cavities.

As noted above, HfO₂ was the first successful application of ALD in semiconductor manufacturing. Its precursors, HfCl₄ and water, are both chemically simple small molecules, whose by-products are volatile and easily removed. Such simple chemistries are the exception, though. ALD of silicon dioxide typically uses aminosilane precursors.⁠[1] Metal nitrides often have complex metal-organic precursor gases. Gassilloud noted that ligands might be added to a precursor molecule to change its vapor pressure or reactivity, or to facilitate adhesion to the substrate. In selective deposition processes, discussed below, ligands might improve selectivity between growth and non-growth surfaces. These larger molecules can be difficult to insinuate into smaller features, and byproducts can be difficult to remove. Complex byproducts can also become a contamination source.

One of the advantages of ALD is its very low process temperature, typically between 200°C and 300°C. It is thermally compatible with both transistor and interconnect processes in CMOS, as well as with deposition on plastic and other novel substrates. Even so, Aditya Kumar and colleagues at GlobalFoundries showed that precise temperature control is important.^⁠[2] TDMAT (tetrakis- dimethylamino titanium) condensation in a TiN deposition process was a significant source of particle defects. To maintain the desired process temperature, both the precursor and purge gas temperatures matter. Introducing cold purge gas into a warm process chamber can cause rapid condensation.

As ALD has become a mainstream process, the industry has found applications for it beyond core device materials, in a variety of sacrificial and spacer layers. For example, double- and quadruple-patterning schemes often use ALD for “pitch-doubling.” By depositing a spacer material on either side of a patterned “mandrel,” then removing the mandrel, the process can cut the original pitch in half without the need for an additional, more costly lithography step.^⁠[3]

Fig. 1: Self-aligned double patterning with ALD spacers. Source: Creative Commons

Depositing a doped oxide on the vertical silicon fins of a finFET device is a less directional and less damaging alternative to ion implantation.[^⁠4]

Selective deposition brings lateral control
These last two examples depend on surface characteristics to mediate deposition. A precursor might adhere more readily to a hard mask than to the underlying material. The vertical face of a silicon fin might offer more (or fewer) adsorption sites than the horizontal face. Selective deposition on more complicated structures may require a pre-deposited growth template, functionalizing substrate regions to encourage or discourage growth. Selective deposition is especially important in interconnect applications. In general, though, a comprehensive review by Rong Chen and colleagues at Huazhong University of Science and Technology explained that selective deposition methods need to replenish the template material as the film grows while needing a mechanism to selectively remove the unwanted material.⁠[5]

For example, tungsten preferentially deposits on silicon relative to SiO₂, but the selectivity diminishes after only a few cycles. Researchers at North Carolina State University successfully re-passivated the oxide by incorporating hydrogen into the tungsten precursor.[⁠6] Similarly, a group at Argonne National Laboratory found that SiO₂ preferentially deposited on SiO₂ relative to other oxides for only 10 to 15 cycles. Adding acetylacetone (“Precursor C”) as an inhibitor every 5 to 10 cycles — restored selectivity.⁠[7]

Alternatively, or in addition, atomic layer etching (ALE) might be used to remove unwanted material. ALE operates in the same step-by-step manner as ALD. The first half of a cycle reacts with the existing surface, weakening the bond to the underlying material. Then, a second step — typically ion bombardment — removes the weakened layer. For example, in ALE etching of silicon, chlorine gas reacts with the surface to form various SiCl_x compounds. The chlorination process weakens the inter-silicon bonds between the surface and the bulk, and the chlorinated layer is easily sputtered away. The layer-by-layer nature of ALE depends on preferential removal of the surface material relative to the bulk (SiCl_x vs. Si in this case). The “ALE window” is the combination of energy and temperature at which the surface layer is completely removed without damaging the underlying material.

Somewhat counter-intuitively, Keren Kanarik and colleagues at Lam Research found that higher ion energies actually expanded the ALE window for silicon etching. High ion energies with short exposure times delayed the onset of silicon sputtering relative to conventional RIE.[8]

Adding and subtracting, one atomic layer at a time
For a long time, the semiconductor industry has been looking for alternatives to process schemes that deposit material, pattern it, then etch most of it away. Wouldn’t it be simpler to only deposit the material we will ultimately need? Meanwhile, atomic layer deposition has been filling the spaces under nanosheets and inside cavities. Bulk deposition and etch tools are still with us, and will be for the foreseeable future. In more and more cases, though, those tools provide the frame while ALD and ALE processes fill in the details.

References

1. Wenling Li, et al., “Impact of aminosilane and silanol precursor structure on atomic layer deposition process,”Applied Surface Science, Vol 621, 2023,156869, https://doi.org/10.1016/j.apsusc.2023.156869.
2. Kumar, et al., “ALD TiN Surface Defect Reduction for 12nm and Beyond Technologies,” 2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC), Saratoga Springs, NY, USA, 2020, pp. 1-4, doi: 10.1109/ASMC49169.2020.9185271.
3. Shohei Yamauchi, et al., “Extendibility of self-aligned type multiple patterning for further scaling”, Proc. SPIE 8682, Advances in Resist Materials and Processing Technology XXX, 86821D (29 March 2013); https://doi.org/10.1117/12.2011953
4. Kalkofen, et al., “Atomic layer deposition of phosphorus oxide films as solid sources for doping of semiconductor structures,” 2018 IEEE 18th International Conference on Nanotechnology (IEEE-NANO), Cork, Ireland, 2018, pp. 1-4, doi: 10.1109/NANO.2018.8626235.
5. Rong Chen et al., “Atomic level deposition to extend Moore’s law and beyond,” 2020 Int. J. Extrem. Manuf. 2 022002 DOI 10.1088/2631-7990/ab83e0
6. B Kalanyan, et al., “Using hydrogen to expand the inherent substrate selectivity window during tungsten atomic layer deposition,” 2016 Chem. Mater. 28 117–26 https://doi.org/10.1021/acs.chemmater.5b03319
7. Yanguas-Gil A, Libera J A and Elam J W, “Modulation of the growth per cycle in atomic layer deposition using reversible surface functionalization,” 2013 Chem. Mater. 25 4849–60 https://doi.org/10.1021/cm4029098
8. Keren J. Kanarik, et al., “Universal scaling relationship for atomic layer etching,” J. Vac. Sci. Technol. A 39, 010401 (2021); doi: 10.1116/6.0000762

The post Enabling Advanced Devices With Atomic Layer Processes appeared first on Semiconductor Engineering.

Successive versions of vertical transistors are emerging as the likely successor to finFETs, combining lower leakage with significant area reduction.

A stacked nanosheet transistor, introduced at N3, uses multiple channel layers to maintain the overall channel length and necessary drive current while continuing to reduce the standard cell footprint. The follow-on technology, the CFET, pushes further up the z axis, stacking n-channel and p-channel transistors on top of each other, rather than side by side.

In work presented at December’s IEEE Electron Device Meeting, researchers at TSMC estimated that CFETs give a 1.5X to 2X overall size reduction at constant gate dimensions. [1] Those are significant area benefits for any digital logic, but manufacturing these new transistor structures will be a challenge.

Monolithic 3D integration is the simplest integration scheme, and the one likely to see production first. In monolithic 3D integration, the entire structure is assembled on a single piece of silicon. This approach can also be used to fabricate compute-in-memory designs where memory devices are fabricated as part of the metallization layers for a conventional CMOS circuit. While individual layers in monolithic 3D designs can incorporate new technologies — the integration of ReRAM devices, for example — the overall CMOS flow is preserved. All of the materials and processes used must be compatible with that rubric.

Adding more nanosheets for complementary devices
The overall process in this kind of scheme is similar to a stacked nanosheet transistor flow. It starts with a stack of eight or more alternating silicon and silicon germanium layers (four pairs), compared to a stacked nanosheet NFET or PFET, which might have only four such layers (two pairs). In a CFET flow, however, middle dielectric layer is inserted halfway through the stack.

This layer, separating the n-type and p-type transistors, is probably the most important difference from a standard nanosheet transistor flow. To minimize parasitic capacitance, the middle dielectric layer should be as thin as possible, said imec’s Naoto Horiguchi. If it’s too thin, though, edge placement errors can cause isolation failures, landing contacts for the top devices onto bottom devices. [2]

In TSMC’s process, the Si/SiGe superlattice includes a high-germanium SiGe layer as a placeholder for the middle dielectric. After the source/drain etch, a highly selective etch removes this layer and oxidizes the silicon on either side of it to form the middle dielectric.

The inner spacer recess etch, which follows middle dielectric formation in the TSMC process, indents the SiGe layers relative to the silicon nanosheets, defining the gate length and junction overlap.

While TSMC emphasized it has not yet made fully metallized integrated CFET circuits, it did report that more than 90% of the transistors survived.

Fig. 1: TSMC used monolithic integration to stack NFET and PFET devices. [1]

Depositing the nanosheet stack is straightforward. Etching it with the precision required is not. A less-than-vertical etch profile will change the relative channel lengths of the top and bottom devices, leading to asymmetric switching characteristics.

Stacking wafers for more flexibility
The alternative, sequential 3D integration is a bit more flexible. While monolithic 3D integration uses a single device layer, sequential 3D integration bonds an additional tier on top of the first. Sequential 3D integration is different from three-dimensional wafer-level packaging and chip stacking, though. In WLP, the component devices are finished, passivated, and tested. The component chips are fully functional as independent circuits. In sequential 3D integration, the two tiers are part of a single integrated circuit.

Often, though not always, the second tier is an unprocessed bare wafer with no devices at all. Ionut Radu, director of research and external collaborations at Soitec, said his company used its SmartCut process to transfer sub-micron silicon layers. [3] One of the advantages of sequential integration, though, is that it opens the door to other possibilities. For example, the second layer could use a different silicon lattice orientation to facilitate stress engineering for improved carrier mobility. It also could use an alternative channel material, such as GaAs or a two-dimensional semiconductor. And up until the transfer occurs, processing of the second wafer has no effect on the thermal budget of the first.

After bonding, the second tier’s process temperature generally must remain below 500° C. Tadeu Mota-Frutuoso, process integration engineer at CEA-Leti, said researchers were able to achieve this benchmark in a conventional CMOS process by using laser annealing for the source/drain activation steps. [4]

While sequential 3D integration can be used to realize CFET devices, the top layers also can contain independent circuitry. Still, as in monolithic integration, the dielectric layer between the two circuit tiers is a critical process step. Analysts at KAIST found that reducing the thickness of the interlayer dielectric improves heat dissipation. It also facilitates the use of a bottom gate to control the top tier devices. On the other hand, the dielectric layer lies at the interface between the original wafer and the transferred layer. Thickness control depends on the polishing step used to prepare the transfer surface. Such precise control is extremely challenging for CMP. [5]

Re-driving wafers without contamination
While the second circuit tier can be added at any point in the process flow, the insertion point constrains not only the first and second tier devices, but also the fab as a whole. When the second layer does not yet contain devices, alignment to the first layer is relatively easy. In contrast, Horiguchi said, aligning one device wafer on top of another imposes an area penalty to accommodate potential overlay error. The total device thickness of sequential 3D structures tends to be greater, as well.

Returning a first-tier wafer with contacts and other metallization to FEOL tools for fabrication of a second transistor layer poses a substantial cross contamination risk. Even if the top surface is well encapsulated, Mota-Frutuoso explained in an interview that the sidewalls and bevels of the bottom tier can still expose metal layers to FEOL processes. CEA-Leti’s proposed bevel contamination wrap (BCW) scheme first cleans the wafer edge, then encapsulates it and the sidewall in a protective oxide layer.

Fig. 2: CEA-Leti’s sequential 3D integration stacked silicon CMOS on an industrial 28nm FDSOI wafer. [4]

Controlling heat dissipation
Heat dissipation is a major challenge for both monolithic and sequential 3D devices. Generalizations are difficult because thermal characteristics depend on the specific integration scheme and even the circuit design. Wei-Yen Woon, senior manager at TSMC, and his colleagues evaluated AlN and diamond as possible thermal dissipation layers. While both have been used in power devices, they are new to CMOS process flows. They achieved good quality columnar AlN films with a low temperature sputtering process, though the columnar structure did impede in-plane heat transport. While diamond offers extremely high thermal conductivity, it also can require extremely high process temperatures. The TSMC group deposited thin films with acceptable quality at BEOL compatible temperatures by using pre-deposited diamond nuclei, but they have not yet attempted to integrate these films with working devices.^[6]

What’s next?
In the short term, monolithic 3D integration offers a relatively straightforward path to CFET fabrication, building on existing nanosheet transistor process flows. Even proponents of sequential 3D integration expect the monolithic approach to reach production first. For the longer term, though, the ability to use a completely different material for the second device layer gives device designers many more process optimization knobs.

However it is achieved, the idea that active devices no longer need to confine themselves to a single planar layer has implications far beyond logic transistors. From compute-in-memory modules to image sensors, 3D integration is an important tool for “More than Moore” devices.

References

^[1] S. Liao et al., “Complementary Field-Effect Transistor (CFET) Demonstration at 48nm Gate Pitch for Future Logic Technology Scaling,” 2023 International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 2023, pp. 1-4, doi: 10.1109/IEDM45741.2023.10413672.

^[2] N. Horiguchi et al., “3D Stacked Devices and MOL Innovations for Post-Nanosheet CMOS Scaling,” 2023 International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 2023, pp. 1-4, doi: 10.1109/IEDM45741.2023.10413701.

^[3] I. Radu et al., “Ultimate Layer Stacking Technology for High Density Sequential 3D Integration,” 2023 International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 2023, pp. 1-4, doi: 10.1109/IEDM45741.2023.10413807.

^[4] T. Mota-Frutuoso et al., “3D sequential integration with Si CMOS stacked on 28nm industrial FDSOI with Cu-ULK iBEOL featuring RO and HDR pixel,” 2023 International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 2023, pp. 1-4, doi: 10.1109/IEDM45741.2023.10413864.

^[5] S. K. Kim et al., “Role of Inter-Layer Dielectric on the Electrical and Heat Dissipation Characteristics in the Heterogeneous 3D Sequential CFETs with Ge p-FETs on Si n-FETs,” 2023 International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 2023, pp. 1-4, doi: 10.1109/IEDM45741.2023.10413845.

^[6] W. Y. Woon et al., “Thermal dissipation in stacked devices,” 2023 International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 2023, pp. 1-4, doi: 10.1109/IEDM45741.2023.10413721.

The post Building CFETs With Monolithic And Sequential 3D appeared first on Semiconductor Engineering.

In an exclusive interview ahead of an invite-only event today in San Jose, Intel outlined new chip technologies it will offer its foundry customers by sharing a glimpse into its future data-center processors. The advances include more dense logic and a 16-fold increase in the connectivity within 3D-stacked chips, and they will be among the first top-end technologies the company has ever shared with chip architects from other companies.

The new technologies will arrive at the culmination of a years-long transformation for Intel. The processor maker is moving from being a company that produces only its own chips to becoming a foundry, making chips for others and considering its own product teams as just another customer. The San Jose event, IFS Direct Connect, is meant as a sort of coming-out party for the new business model.

Internally, Intel plans to use the combination of technologies in a server CPU code-named Clearwater Forest. The company considers the product, a system-on-a-chip with hundreds of billions of transistors, an example of what other customers of its foundry business will be able to achieve.

“Our objective is to get the compute to the best performance per watt we can achieve” from Clearwater Forest, said Eric Fetzer, director of data center technology and pathfinding at Intel. That means using the company’s most advanced fabrication technology available, Intel 18A.

3D stacking “improves the latency between compute and memory by shortening the hops, while at the same time enabling a larger cache” —Pushkar Ranade

“However, if we apply that technology throughout the entire system, you run into other potential problems,” he added. “Certain parts of the system don’t necessarily scale as well as others. Logic typically scales generation to generation very well with Moore’s Law.” But other features do not. SRAM, a CPU’s cache memory, has been lagging logic, for example. And the I/O circuits that connect a processor to the rest of a computer are even further behind.

Faced with these realities, as all makers of leading-edge processors are now, Intel broke Clearwater Forest’s system down into its core functions, chose the best-fit technology to build each, and stitched them back together using a suite of new technical tricks. The result is a CPU architecture capable of scaling to as many as 300 billion transistors.

In Clearwater Forest, billions of transistors are divided among three different types of silicon ICs, called dies or chiplets, interconnected and packaged together. The heart of the system is as many as 12 processor-core chiplets built using the Intel 18A process. These chiplets are 3D-stacked atop three “base dies” built using Intel 3, the process that makes compute cores for the Sierra Forest CPU, due out this year. Housed on the base die will be the CPU’s main cache memory, voltage regulators, and internal network. “The stacking improves the latency between compute and memory by shortening the hops, while at the same time enabling a larger cache,” says senior principal engineer Pushkar Ranade.

Finally, the CPU’s I/O system will be on two dies built using Intel 7, which in 2025 will be trailing the company’s most advanced process by a full four generations. In fact, the chiplets are basically the same as those going into the Sierra Forest and Granite Rapids CPUs, lessening the development expense.

Here’s a look at the new technologies involved and what they offer:

3D Hybrid Bonding

3D hybrid bonding links compute dies to base dies.Intel

Intel’s current chip-stacking interconnect technology, Foveros, links one die to another using a vastly scaled-down version of how dies have long been connected to their packages: tiny “microbumps” of solder that are briefly melted to join the chips. This lets today’s version of Foveros, which is used in the Meteor Lake CPU, make one connection roughly every 36 micrometers. Clearwater Forest will use new technology, Foveros Direct 3D, which departs from solder-based methods to bring a whopping 16-fold increase in the density of 3D connections.

Called “hybrid bonding,” it’s analogous to welding together the copper pads at the face of two chips. These pads are slightly recessed and surround by insulator. The insulator on one chip affixes to the other when they are pressed together. Then the stacked chips are heated, causing the copper to expand across the gap and bind together to form a permanent link. Competitor TSMC uses a version of hybrid bonding in certain AMD CPUs to connect extra cache memory to processor-core chiplets and, in AMD’s newest GPU, to link compute chiplets to the system’s base die.

“The hybrid bond interconnects enable a substantial increase in density” of connections, says Fetzer. “That density is very important for the server market, particularly because the density drives a very low picojoule-per-bit communication.” The energy involved in data crossing from one silicon die to another can easily consume a big chunk of a product’s power budget if the per-bit energy cost is too high. Foveros Direct 3D brings that cost down below 0.05 picojoules per bit, which puts it on the same scale as the energy needed to move bits around within a silicon die.

A lot of that energy savings comes from the data traversing less copper. Say you wanted to connect a 512-wire bus on one die to the same-size bus on another so the two dies can share a coherent set of information. On each chip, these buses might be as narrow as 10–20 wires per micrometer. To get that from one die to the other using today’s 36-micrometer-pitch microbump tech would mean scattering those signals across several hundred square micrometers of silicon on one side and then gathering them across the same area on the other. Charging up all that extra copper and solder “quickly becomes both a latency and a large power problem,” says Fetzer. Hybrid bonding, in contrast, could do the bus-to-bus connection in the same area that a few microbumps would occupy.

As great as those benefits might be, making the switch to hybrid bonding isn’t easy. To forge hybrid bonds requires linking an already-diced silicon die to one that’s still attached to its wafer. Aligning all the connections properly means the chip must be diced to much greater tolerances than is needed for microbump technologies. Repair and recovery, too, require different technologies. Even the predominant way connections fail is different, says Fetzer. With microbumps, you are more likely to get a short from one bit of solder connecting to a neighbor. But with hybrid bonding, the danger is defects that lead to open connections.

Backside power

One of the main distinctions the company is bringing to chipmaking this year with its Intel 20A process, the one that will precede Intel 18A, is backside power delivery. In processors today, all interconnects, whether they’re carrying power or data, are constructed on the “front side” of the chip, above the silicon substrate. Foveros and other 3D-chip-stacking tech require through-silicon vias, interconnects that drill down through the silicon to make connections from the other side. But back-side power delivery goes much further. It puts all of the power interconnects beneath the silicon, essentially sandwiching the layer containing the transistors between two sets of interconnects.

PowerVia puts the silicon’s power supply network below, leaving more room for data-carrying interconnects above.Intel

This arrangement makes a difference because power interconnects and data interconnects require different features. Power interconnects need to be wide to reduce resistance, while data interconnects should be narrow so they can be densely packed. Intel is set to be the first chipmaker to introduce back-side power delivery in a commercial chip, later this year with the release of the Arrow Lake CPU. Data released last summer by Intel showed that back-side power alone delivered a 6 percent performance boost.

The Intel 18A process technology’s back-side-power-delivery network technology will be fundamentally the same as what’s found in Intel 20A chips. However, it’s being used to greater advantage in Clearwater Forest. The upcoming CPU includes what’s called an “on-die voltage regulator” within the base die. Having the voltage regulation close to the logic it drives means the logic can run faster. The shorter distances let the regulator respond to changes in the demand for current more quickly, while consuming less power.

Because the logic dies use back-side power delivery, the resistance of the connection between the voltage regulator and the dies logic is that much lower. “The power via technology along with the Foveros stacking gives us a really efficient way to hook it up,” says Fetzer.

RibbonFET, the next generation

In addition to back-side power, the chipmaker is switching to a different transistor architecture with the Intel 20A process: RibbonFET. A form of nanosheet, or gate-all-around, transistor, RibbonFET replaces the FinFET, CMOS’s workhorse transistor since 2011. With Intel 18A, Clearwater Forest’s logic dies will be made with a second generation of RibbonFET process. While the devices themselves aren’t very different from the ones that will emerge from Intel 20A, there’s more flexibility to the design of the devices, says Fetzer.

RibbonFET is Intel’s take on nanowire transistors.Intel

“There’s a broader array of devices to support various foundry applications beyond just what was needed to enable a high-performance CPU,” which was what the Intel 20A process was designed for, he says.

RibbonFET’s nanowires can have different widths depending on the needs of a logic cell.Intel

Some of that variation stems from a degree of flexibility that was lost in the FinFET era. Before FinFETs arrived, transistors in the same process could be made in a range of widths, allowing a more-or-less continuous trade-off between performance—which came with higher current—and efficiency—which required better control over leakage current. Because the main part of a FinFET is a vertical silicon fin of a defined height and width, that trade-off now had to take the form of how many fins a device had. So, with two fins you could double current, but there was no way to increase it by 25 or 50 percent.

With nanosheet devices, the ability to vary transistor widths is back. “RibbonFET technology enables different sizes of ribbon within the same technology base,” says Fetzer. “When we go from Intel 20A to Intel 18A, we offer more flexibility in transistor sizing.”

That flexibility means that standard cells, basic logic blocks designers can use to build their systems, can contain transistors with different properties. And that enabled Intel to develop an “enhanced library” that includes standard cells that are smaller, better performing, or more efficient than those of the Intel 20A process.

2nd generation EMIB

In Clearwater Forest, the dies that handle input and output connect horizontally to the base dies—the ones with the cache memory and network—using the second generation of Intel’s EMIB. EMIB is a small piece of silicon containing a dense set of interconnects and microbumps designed to connect one die to another in the same plane. The silicon is embedded in the package itself to form a bridge between dies.

Dense 2D connections are formed by a small sliver of silicon called EMIB, which is embedded in the package substrate.Intel

The technology has been in commercial use in Intel CPUs since Sapphire Rapids was released in 2023. It’s meant as a less costly alternative to putting all the dies on a silicon interposer, a slice of silicon patterned with interconnects that is large enough for all of the system’s dies to sit on. Apart from the cost of the material, silicon interposers can be expensive to build, because they are usually several times larger than what standard silicon processes are designed to make.

The second generation of EMIB debuts this year with the Granite Rapids CPU, and it involves shrinking the pitch of microbump connections from 55 micrometers to 45 micrometers as well as boosting the density of the wires. The main challenge with such connections is that the package and the silicon expand at different rates when they heat up. This phenomenon could lead to warpage that breaks connections.

What’s more, in the case of Clearwater Forest “there were also some unique challenges, because we’re connecting EMIB on a regular die to EMIB on a Foveros Direct 3D base die and a stack,” says Fetzer. This situation, recently rechristened EMIB 3.5 technology (formerly called co-EMIB), requires special steps to ensure that the stresses and strains involved are compatible with the silicon in the Foveros stack, which is thinner than ordinary chips, he says.

For more, see Intel’s whitepaper on their foundry tech.