Semiconductor Energy Laboratory Co., Ltd.

AI has been rapidly developed in recent years, and we easily use AI with PCs or smartphones. However, AI needs the enormous amount of arithmetic processing and thus consumes a lot of power. Such global power consumption increase caused by the spread of AI is now recognized as an urgent issue.
SEL has been developing OSLSI^® technology for AI to solve the issue. An oxide semiconductor (OS) transistor has extremely low leakage current and thus enables a memory holding data at high accuracy. We utilize such features to realize AI with low power consumption.

AiMC AI chip Brainmorphic Computer

In-memory computing¹⁾ (iMC) has attracted attention as an architecture of improving efficiency of AI arithmetic processing. SEL develops AI chips using analog in-memory computing (AiMC) to which OSLSI technology is applied. In NOSRAM^®, oxide semiconductor (OS) transistors can hold analog data and silicon (Si) transistors can operate at low voltage; thus, arithmetic processing can be performed at a minute current.
Devising a method for driving a memory, a method for correcting element variations, etc., enable many memories to be utilized in parallel, which leads to efficient arithmetic processing. We fabricated and demonstrated AI chips to confirm that arithmetic processing can be performed high efficiently compared with conventional techniques^{[1], [2], [3]}.




[1] H. Baba et al., “Novel Analog in-Memory Compute with < 1 nA Current/Cell and 143.9 TOPS/W Enabled by Monolithic Normally-off Zn-rich CAAC-IGZO FET-on-Si CMOS Technology,” IEDM Tech. Dig., 466 (2021).
[2] M. -C. Chen et al., “A > 64 Multiple States and > 210 TOPS/W High Efficient Computing by Monolithic Si/CAAC-IGZO + Super-Lattice ZrO₂/Al₂O₃/ZrO₂ for Ultra-Low Power Edge AI Application,” IEDM Tech. Dig., 423 (2022).
[3] H. Rikimaru et al., “Crystalline Oxide Semiconductor FET-based Analog Neural Network for Intelligent IoT Sensor,” Ext. Abstr. Solid State Devices and Materials, 772 (2022).

AI × Image Sensor

We develop image sensors with AI image processing inside as one of applications of OSLSI technology. For example, such an image sensor is incorporated in a car to recognize people, obstacles, signs, etc., by AI image processing, which can improve safety while driving.

BTOS^® (Batteries Oxide Semiconductor/ Operating System)

Micro short-circuit of a lithium-ion battery cell is well known to be a major cause for ignition accidents. As one of applications of OSLSIs, we develop BTOS technology^[5] that controls batteries with AI. The BTOS technology detects micro short-circuit by comparing voltage of a lithium ion battery with a reference voltage using a micro short-circuit detector that includes an OSLSI memory and an OSLSI comparator. BTOS is suitably mounted in a battery pack because of its simple structure.


[4] Demonstrated at AI EXPO 2017, June 28 to 30, 2017, Tokyo Big Sight, Japan.
[5] H. Inoue et al., “Micro Short-Circuit Detector Including S/H Circuit for 1hr Retention and 52dB Comparator Composed of C-Axis Aligned Crystalline IGZO FETs for Li-Ion Battery Protection IC,” Int. Solid-State Circuits Conf. Dig. Tech. Pap., 204 (2019).

Idling-Stop (IDS) Driving

As one of applications of OSLSI technology, SEL develops controlling IDS driving with AI. This technology predicts a period in which an image of a display does not change, that is, a period that does not need refreshing (re-writing of an image), with AI. IDS driving can be effectively performed, which enables ultra-low power consumption of displays^[6].


[6] H. Kunitake et al., “Low-Power Display System Enabled by Combining Oxide-Semiconductor and Neural-Network Technologies,” ECS Trans., 79(1), 177 (2017).

* OSLSI, Brainmorphic Computer, NOSRAM, BTOS, and CAAC-IGZO are registered trademarks of Semiconductor Energy Laboratory Co., Ltd.
(Japanese trademark registration No. 5698906, No. 6130966, No. 5529056, No. 6146374, and No. 5494218).