In this work, we developed a nonvolatile ternary content addressable memory (TCAM) with a cell size of $0.01~\mu \text{m}^{{2}}$ utilizing the Ge-based memory diode (MD), which has the most area-efficient TCAM design reported. The MDs have a high current ratio between ON and OFF states and a large rectifying ratio, showing the potential usage in large-dimension TCAM arrays. Besides, the functionality of parallel search was demonstrated with a 2-bit MD-TCAM array by experiment, and the electrical characterization showed expected results. With the help of the sub-ns ultrafast measurement system, it is confirmed that the search energy of MD-TCAM could reach as low as 1.0 fJ/bit/mismatch, and one search operation can be performed within 200 ps. Furthermore, the circuit-level simulation results verified that the MD-TCAM developed in this study shows good performance in 128-bit parallel searching, which is promising for the ultrafast and low-power data search applications in the coming IoT era.
This work presents the first demonstration of the transistor-free memory-diode (MD) array fabricated on Ge-on-Insulator (GeOI), which can work as an ultra-compact 8 F 2 ternary content addressable memory (TCAM). Through the wafer bonding and thinning techniques, the Ge layer in the GeOI structure maintains the monocrystalline quality originated from bulk Ge to ensure the Fermi-level pinning effect, forming an effective Schottky barrier with low reverse-bias leakage currents. Furthermore, an ultra-thin Y-doped GeO x layer is introduced to suppress the surface-state current, enabling a higher rectifying ratio. Thanks to the ultra-high rectifying (> 2 × 10 5 ) and sufficient on/off ratios (> 500), the 4-bits MD-TCAM array shows over 2 orders differences between the match and mismatch currents, which experimentally verifies the feasibility and accuracy of parallel data search.
11 October 2024 in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
This work presents an analog computing-in-memory (CiM) macro with STT-MRAM and 28 nm CMOS technology. The adopted CiM bitcell uses a differential scheme and balances the input resistance to minimize the non-ideal factors during multiply-accumulate (MAC) operations. Specialized peripheral circuits were designed for the current-scheme CiM architecture. More importantly, strategies of accuracy improvement were innovatively proposed: (1) Mapping most-significant-bit (MSB) to the far side of the MRAM array, (2) Output linear transformation based on the reference columns. Circuit-level simulation verified the functionality and performance improvement of the CiM macro based on the MNIST and CIFAR-10 datasets, realizing a 3% and 5% accuracy loss compared with the benchmark respectively. The 640 GOPS (8-bit) throughput, 34.6 TOPS/mm2 area compactness and 83.3 TOPS/W energy efficiency demonstrate the advantages of STT-MRAM CiM in the coming AI-era.