3DVLSI with CoolCube process: An alternative path to scaling

Abstract

3D VLSI with a CoolCube TM integration allows vertically stacking several layers of devices with a unique connecting via density above a million/mm². This results in increased density with no extra cost associated to transistor scaling, while benefiting from gains in power and performance thanks to wire-length reduction. CoolCube TM technology leads to high performance top transistors with Thermal Budgets (TB) compatible with bottom MOSFET integrity. Key enablers are the dopant activation by Solid Phase Epitaxy (SPE) or nanosecond laser anneal, low temperature epitaxy, low k spacers and direct bonding. New data on the maximal TB bottom MOSFET can withstand (with high temperatures but short durations) offer new opportunities for top MOSFET process optimization. I-INTRODUCTION: 3D-VLSI sequential processing of stacked devices offers a 3D contact pitch equal to planar contacted gate pitch. Moreover, thanks to the lithography alignment precision depending only on the stacked technology node, the via density reaches 10 8 vias/mm 2 with a 90nm-CPP technology. This 3D contact pitch allows fully leveraging the 3 rd dimension with negligible area penalty due to 3D via or landing pad size compared to packaging solutions as shown in Fig.1 [1] .3DVLSI motivations-Actually, scaling below the 28 nm technology node does not yield substantial cost reductions [2] . Among scaling challenges, the increasing interconnect delay overshadows benefits stemming from costly transistor scaling. 3D VLSI interest has been demonstrated via a Power-Performance-Area benchmark for FPGA applications stacking two layers of 14nm FDSOI technology with W/SiO2 interconnect in between [3] . Dramatic area and Energy Delay Product (EDP) reduction is achieved, with benefits exceeding those of downscaling to the 10 nm node (Figs.2&3). 3DVLSI partitioning at the gate level allows IC performance gain without resorting to scaling thanks to wire length reduction. In parallel, partitioning at the transistor level by stacking n-FET over p-FET (or the opposite) enables the independent optimization of both types of transistors (customized implementation of performance boosters: channel material / substrate orientation / channel and Raised Sources and Drains strain, etc. [6,7] ) with reduced process complexity compared to a planar co-integration. The ultimate example of high performance CMOS at low process cost is the stacking of III-V nFETs above SiGe pFETs [8,9] . These high mobility transistors are well suited for 3DVLSI because their process temperatures are intrinsically low. 3DVLSI partitioning at the transistor level allows performance gain as it facilitates the cointegration of high performance n&p-FETs compared to a coplanar scheme. Finally, 3DVLSI, with its high contact density, is also anticipated as a powerful solution for heterogeneous cointegrations requiring high 3D vias densities such as NEMS with CMOS for gas sensing applications [10, 11] or highly miniaturized imagers [12] . Fig.4 summarizes the three main integration schemes foreseen for ...

Monolithically Integrated CMOS-Compatible III–V on Silicon Lasers

“…As our lasers are integrated before processing the BEOL, only the underlying (Bi-)CMOS performance could be affected by this process step. However, stable FET performance has been demonstrated with this thermal budget [29] and thus CMOS compatibility is ensured throughout this process. The devices were finally capped with 800 nm of PECVD-deposited SiO 2 , vias were etched down to the contact metals and a second metal layer was deposited by sputtering and patterned by dry etching.…”

Energy-Efficient 4T-based SRAM Bitcell for Ultra-Low-Voltage Operations in 28nm 3D CoolCubeTM Technology

“…Among the 3D monolithic technologies, 3D sequential LETI CoolCube TM technology [10] offers a fine 3D interconnect pitch compared to existing technologies. This feature the way for efficient 3D-VLSI circuits, aiming to reduce the congestion on the BEOL while providing real 3D routing possibilities [11].…”

Toward Human-Scale Brain Computing Using 3D Wafer Scale Integration

“…Today, multi-stacks of 4 silicon wafers, thinned to 5 μm with 0.25-1 μm TSV features, have been demonstrated , as has functional control of memory on one wafer using logic on another . Wafers may be fabricated in parallel and then bonded or bonded first and then processed sequentially [Batude et al 2015], with different tradeoffs. For this application, we favor the parallel approach because each wafer can be processed and tested independently, leading to higher yield and robustness.…”