Three-dimensional integrated circuit

In microelectronics, a three-dimensional integrated circuit (3D IC) is an integrated circuit manufactured by stacking silicon wafers or dies and interconnecting them vertically using, for instance, through-silicon vias (TSVs) or Cu-Cu connections, so that they behave as a single device to achieve performance improvements at reduced power and smaller footprint than conventional two dimensional processes. 3D IC is just one of a host of 3D integration schemes that exploit the z-direction to achieve electrical performance benefits.

3D integrated circuits can be classified by their level of interconnect hierarchy at the global (package), intermediate (bond pad) and local (transistor) level [1] In general, 3D integration is a broad term that includes such technologies as 3D wafer-level packaging (3DWLP); 2.5D and 3D interposer-based integration; 3D stacked ICs (3D-SICs), monolithic 3D ICs; 3D heterogeneous integration; and 3D systems integration.[2][3]

International organizations such as the Jisso Technology Roadmap Committee (JIC) and the International Technology Roadmap for Semiconductors (ITRS) have worked to classify the various 3D integration technologies to further the establishment of standards and roadmaps of 3D integration.[4]


3D ICs vs. 3D packaging

3D Packaging refers to 3D integration schemes that rely on traditional methods of interconnect such as wire bonding and flip chip to achieve vertical stacks. 3D packaging can be disseminated further into 3D system in package (3D SiP) and 3D wafer level package (3D WLP). Stacked memory die interconnected with wire bonds, and package on package (PoP) configurations interconnected with either wire bonds, or flip chips are 3D SiPs that have been in mainstream manufacturing for some time and have a well established infrastructure. PoP is used for vertically integrating disparate technologies such as 3D WLP uses wafer level processes such as redistribution layers (RDL) and wafer bumping processes to form interconnects.

2.5D interposer is also a 3D WLP that interconnects die side-side on a silicon, glass or organic interposer using TSVs and RDL. In all types of 3D Packaging, chips in the package communicate using off-chip signaling, much as if they were mounted in separate packages on a normal circuit board.

3D ICs can be divided into 3D Stacked ICs (3D SIC), which refers to stacking IC chips using TSV interconnects, and monolithic 3D ICs, which use fab processes to realize 3D interconnects at the local levels of the on-chip wiring hierarchy as set forth by the ITRS, this results in direct vertical interconnects between device layers. The first examples of a monolithic approach are seen in Samsung’s 3D VNAND devices.

One master die and three slave dies

3D SiCs

The digital electronics market requires a higher density semiconductor memory chip to cater to recently released CPU components, and the multiple die stacking technique has been suggested as a solution to this problem. JEDEC disclosed the upcoming DRAM technology includes the "3D SiC" die stacking plan at "Server Memory Forum", November 1–2, 2011, Santa Clara, CA. In August 2014, Samsung started producing 64GB DRAM modules for servers based on emerging DDR4 (double-data rate 4) memory using 3D TSV package technology.[5] Newer proposed standards for 3D stacked DRAM include Wide I/O, Wide I/O 2, Hybrid Memory Cube, High Bandwidth Memory.

Monolithic 3D ICs

Monolithic 3D ICs are built in layers on a single semiconductor wafer, which is then diced into 3D ICs. There is only one substrate, hence no need for aligning, thinning, bonding, or through-silicon vias. Process temperature limitations are addressed by partitioning the transistor fabrication to two phases. A high temperature phase which is done before layer transfer follow by a layer transfer use ion-cut, also known as layer transfer, which has been used to produce Silicon on Insulator (SOI) wafers for the past two decades. Multiple thin (10s–100s nanometer scale) layers of virtually defect-free Silicon can be created by utilizing low temperature (<400℃) bond and cleave techniques, and placed on top of active transistor circuitry. Follow by finalizing the transistors using etch and deposition processes. This monolithic 3D IC technology has been researched at Stanford University under a DARPA-sponsored grant.

CEA-Leti is also developing monolithic 3D IC approaches, called sequential 3D IC. In 2014, the French research institute introduced its CoolCube™, a low-temperature process flow that provides a true path to 3DVLSI.[6] At Stanford University, researchers are designing monolithic 3D ICs using carbon nanotube (CNT) structures vs. silicon using a wafer-scale low temperature CNT transfer processes that can be done at 120℃.[7]

In general, monolithic 3D ICs are still a developing technology and are considered by most to be several years away from production.

Manufacturing technologies for 3D SiCs

As of 2014, a number of memory products such as High Bandwidth Memory (HBM) and the Hybrid Memory Cube have been launched that implement 3D IC stacking with TSVs. There are a number of key stacking approaches being implemented and explored. These include die-to-die, die-to-wafer, and wafer-to-wafer.

Electronic components are built on multiple die, which are then aligned and bonded. Thinning and TSV creation may be done before or after bonding. One advantage of die-to-die is that each component die can be tested first, so that one bad die does not ruin an entire stack. [8] Moreover, each die in the 3D IC can be binned beforehand, so that they can be mixed and matched to optimize power consumption and performance (e.g. matching multiple dice from the low power process corner for a mobile application).
Electronic components are built on two semiconductor wafers. One wafer is diced; the singulated dice are aligned and bonded onto die sites of the second wafer. As in the wafer-on-wafer method, thinning and TSV creation are performed either before or after bonding. Additional die may be added to the stacks before dicing.
Electronic components are built on two or more semiconductor wafers, which are then aligned, bonded, and diced into 3D ICs. Each wafer may be thinned before or after bonding. Vertical connections are either built into the wafers before bonding or else created in the stack after bonding. These " through-silicon vias" (TSVs) pass through the silicon substrate(s) between active layers and/or between an active layer and an external bond pad. Wafer-to-wafer bonding can reduce yields, since if any 1 of N chips in a 3D IC are defective, the entire 3D IC will be defective. Moreover, the wafers must be the same size, but many exotic materials (e.g. III-Vs) are manufactured on much smaller wafers than CMOS logic or DRAM (typically 300 mm), complicating heterogeneous integration.


While traditional CMOS scaling processes improves signal propagation speed, scaling from current manufacturing and chip-design technologies is becoming more difficult and costly, in part because of power-density constraints, and in part because interconnects do not become faster while transistors do.[9] 3D ICs address the scaling challenge by stacking 2D dies and connecting them in the 3rd dimension. This promises to speed up communication between layered chips, compared to planar layout.[10] 3D ICs promise many significant benefits, including:

More functionality fits into a small space. This extends Moore's law and enables a new generation of tiny but powerful devices.
Partitioning a large chip into multiple smaller dies with 3D stacking can improve the yield and reduce the fabrication cost if individual dies are tested separately. [11] [12]
Heterogeneous integration
Circuit layers can be built with different processes, or even on different types of wafers. This means that components can be optimized to a much greater degree than if they were built together on a single wafer. Moreover, components with incompatible manufacturing could be combined in a single 3D IC. [13] [3]
Shorter interconnect
The average wire length is reduced. Common figures reported by researchers are on the order of 10–15%, but this reduction mostly applies to longer interconnect, which may affect circuit delay by a greater amount. Given that 3D wires have much higher capacitance than conventional in-die wires, circuit delay may or may not improve.
Keeping a signal on-chip can reduce its power consumption by 10–100 times. [14] Shorter wires also reduce power consumption by producing less parasitic capacitance. [15] Reducing the power budget leads to less heat generation, extended battery life, and lower cost of operation.
The vertical dimension adds a higher order of connectivity and offers new design possibilities. [3]
Circuit security
3D integration can achieve security through obscurity; the stacked structure complicates attempts to reverse engineer the circuitry. Sensitive circuits may also be divided among the layers in such a way as to obscure the function of each layer. [16] Moreover, 3D integration allows to integrate dedicated, system monitor-like features in separate layers. [3] The objective here is to implement some kind of hardware firewall for any commodity components/chips to be monitored at runtime, seeking to protect the whole electronic system against run-time attacks as well as malicious hardware modifications.
3D integration allows large numbers of vertical vias between the layers. This allows construction of wide bandwidth buses between functional blocks in different layers. A typical example would be a processor+memory 3D stack, with the cache memory stacked on top of the processor. This arrangement allows a bus much wider than the typical 128 or 256 bits between the cache and processor. [17] Wide buses in turn alleviate the memory wall problem. [18]


Because this technology is new it carries new challenges, including:

While cost is a benefit when compared with scaling, it has also been identified as a challenge to the commercialization of 3D ICs in mainstream consumer applications. However, work is being done to address this. Although 3D technology is new and fairly complex, the cost of the manufacturing process is surprisingly straightforward when broken down into the activities that build up the entire process. By analyzing the combination of activities that lay at the base, cost drivers can be identified. Once the cost drivers are identified, it becomes a less complicated endeavor to determine where the majority of cost comes from and, more importantly, where cost has the potential to be reduced. [19]
Each extra manufacturing step adds a risk for defects. In order for 3D ICs to be commercially viable, defects could be repaired or tolerated, or defect density can be improved. [20] [21]
Heat building up within the stack must be dissipated. This is an inevitable issue as electrical proximity correlates with thermal proximity. Specific thermal hotspots must be more carefully managed.
Design complexity
Taking full advantage of 3D integration requires sophisticated design techniques and new CAD tools. [22]
TSV-introduced overhead
TSVs are large compared to gates and impact floorplans. At the 45 nm technology node, the area footprint of a 10μm x 10μm TSV is comparable to that of about 50 gates. [23] Furthermore, manufacturability demands landing pads and keep-out zones which further increase TSV area footprint. Depending on the technology choices, TSVs block some subset of layout resources. [23] Via-first TSVs are manufactured before metallization, thus occupy the device layer and result in placement obstacles. Via-last TSVs are manufactured after metallization and pass through the chip. Thus, they occupy both the device and metal layers, resulting in placement and routing obstacles. While the usage of TSVs is generally expected to reduce wirelength, this depends on the number of TSVs and their characteristics. [23] Also, the granularity of inter-die partitioning impacts wirelength. It typically decreases for moderate (blocks with 20-100 modules) and coarse (block-level partitioning) granularities, but increases for fine (gate-level partitioning) granularities. [23]
To achieve high overall yield and reduce costs, separate testing of independent dies is essential. [21] [24] However, tight integration between adjacent active layers in 3D ICs entails a significant amount of interconnect between different sections of the same circuit module that were partitioned to different dies. Aside from the massive overhead introduced by required TSVs, sections of such a module, e.g., a multiplier, cannot be independently tested by conventional techniques. This particularly applies to timing-critical paths laid out in 3D.
Lack of standards
There are few standards for TSV-based 3D IC design, manufacturing, and packaging, although this issue is being addressed. [25] [26] In addition, there are many integration options being explored such as via-last, via-first, via-middle; [27] interposers [28] or direct bonding; etc.
Heterogeneous integration supply chain
In heterogeneously integrated systems, the delay of one part from one of the different parts suppliers delays the delivery of the whole product, and so delays the revenue for each of the 3D IC part suppliers.
Lack of clearly defined ownership
It is unclear who should own the 3D IC integration and packaging/assembly. It could be assembly houses like ASE or the product OEMs.

Design styles

Depending on partitioning granularity, different design styles can be distinguished. Gate-level integration faces multiple challenges and currently appears less practical than block-level integration.[29]

Gate-level integration
This style partitions standard cells between multiple dies. It promises wirelength reduction and great flexibility. However, wirelength reduction may be undermined unless modules of certain minimal size are preserved. On the other hand, its adverse effects include the massive number of necessary TSVs for interconnects. This design style requires 3D place-and-route tools, which are unavailable yet. Also, partitioning a design block across multiple dies implies that it cannot be fully tested before die stacking. After die stacking (post-bond testing), a single failed die can render several good dies unusable, undermining yield. This style also amplifies the impact of process variation, especially inter-die variation. In fact, a 3D layout may yield more poorly than the same circuit laid out in 2D, contrary to the original promise of 3D IC integration. [30] Furthermore, this design style requires to redesign available Intellectual Property, since existing IP blocks and EDA tools do not provision for 3D integration.
Block-level integration
This style assigns entire design blocks to separate dies. Design blocks subsume most of the netlist connectivity and are linked by a small number of global interconnects. Therefore, block-level integration promises to reduce TSV overhead. Sophisticated 3D systems combining heterogeneous dies require distinct manufacturing processes at different technology nodes for fast and low-power random logic, several memory types, analog and RF circuits, etc. Block-level integration, which allows separate and optimized manufacturing processes, thus appears crucial for 3D integration. Furthermore, this style might facilitate the transition from current 2D design towards 3D IC design. Basically, 3D-aware tools are only needed for partitioning and thermal analysis. [31] Separate dies will be designed using (adapted) 2D tools and 2D blocks. This is motivated by the broad availability of reliable IP blocks. It is more convenient to use available 2D IP blocks and to place the mandatory TSVs in the unoccupied space between blocks instead of redesigning IP blocks and embedding TSVs. [29] Design-for-testability structures are a key component of IP blocks and can therefore be used to facilitate testing for 3D ICs. Also, critical paths can be mostly embedded within 2D blocks, which limits the impact of TSV and inter-die variation on manufacturing yield. Finally, modern chip design often requires last-minute engineering changes. Restricting the impact of such changes to single dies is essential to limit cost.

Notable 3D chips

In 2004 Tezzaron Semiconductor built working 3D devices from six different designs.[32] The chips were built in two layers with "via-first" tungsten TSVs for vertical interconnection. Two wafers were stacked face-to-face and bonded with a copper process. The top wafer was thinned and the two-wafer stack was then diced into chips. The first chip tested was a simple memory register, but the most notable of the set was an 8051 processor/memory stack that exhibited much higher speed and lower power consumption than an analogous 2D assembly.

In 2004, Intel presented a 3D version of the Pentium 4 CPU.[33] The chip was manufactured with two dies using face-to-face stacking, which allowed a dense via structure. Backside TSVs are used for I/O and power supply. For the 3D floorplan, designers manually arranged functional blocks in each die aiming for power reduction and performance improvement. Splitting large and high-power blocks and careful rearrangement allowed to limit thermal hotspots. The 3D design provides 15% performance improvement (due to eliminated pipeline stages) and 15% power saving (due to eliminated repeaters and reduced wiring) compared to the 2D Pentium 4.

The Teraflops Research Chip introduced in 2007 by Intel is an experimental 80-core design with stacked memory. Due to the high demand for memory bandwidth, a traditional I/O approach would consume 10 to 25 W.[24] To improve upon that, Intel designers implemented a TSV-based memory bus. Each core is connected to one memory tile in the SRAM die with a link that provides 12 GB/s bandwidth, resulting in a total bandwidth of 1 TB/s while consuming only 2.2 W.

An academic implementation of a 3D processor was presented in 2008 at the University of Rochester by Professor Eby Friedman and his students. The chip runs at a 1.4 GHz and it was designed for optimized vertical processing between the stacked chips which gives the 3D processor abilities that the traditional one layered chip could not reach.[34] One challenge in manufacturing of the three-dimensional chip was to make all of the layers work in harmony without any obstacles that would interfere with a piece of information traveling from one layer to another.[35]

In ISSCC 2012, two 3D-IC-based multi-core designs using GlobalFoundries' 130 nm process and Tezzaron's FaStack technology were presented and demonstrated. 3D-MAPS,[36] a 64 custom core implementation with two-logic-die stack was demonstrated by researchers from the School of Electrical and Computer Engineering at Georgia Institute of Technology. The second prototype was from the Department of Electrical Engineering and Computer Science at University of Michigan called Centip3De, a near-threshold design based on ARM Cortex-M3 cores.

Other Languages