innovation marketplace


Discover new technologies by our partners

Leveraging our wide network of partners, we have curated numerous enabling technologies available for licensing and commercialisation across different industries and domains. Enterprises interested in these technology offers and collaborating with partners of complementary technological capabilities can reach out for co-innovation opportunities.

Fault-Tolerant Technique for Deep Learning Accelerator Design
Successful fault-injection attacks on Deep Neural Network (DNN) hardware accelerators for real-world object recognition and classification problems have been demonstrated. For safety and security-critical applications, fault resiliency on computation errors is a desirable property to avert DNN misclassification. Mitigation of computation errors require real-time error detection and correction. Error correction codes are not applicable to arithmetic operations; while doubling or triplicating the functional units with majority voting to achieve fault tolerance is too expensive. Existing lightweight shadow register-based error correction solutions need to stall the execution upon error detection to flush the multiply-accumulate (MAC) pipeline. To recover from the fault, the data will have to be replayed and re-computed. This process introduces non-trivial throughput and power consumption overheads. Other fault resilient DNN hardware designs exploit the inherent redundancy of the DNN model to avoid explicit error correction. However, such design methodologies can only recover the prediction accuracy from uniform errors or sparse faults due to natural noises or particle radiation but not biased and intensive faults arising from deliberate attacks. This technology offer is a hardware design method for DNN convolution operations that provides efficient and timely error correction to prevent prediction accuracy degradation due to both naturally occurring errors as well as maliciously injected faults. Specifically, this design method can be applied to the convolutional layers of any pre-trained DNN models for efficient implementation on both application-specific integrated circuit (ASIC) and field programmable gate array (FPGA) platforms to increase its robustness against fault-injection attacks without impacting the original throughput.
Mutual Authentication and Key Exchange Protocol for Peer-to-Peer IoT Devices
In database-driven Internet of Things (IoT), the connection between two IoT devices is performed indirectly through a server. Data collected on one IoT device needs to flow to the server before reaching another IoT device. By contrast, Peer-to-Peer (P2P) IoT enables direct connection between two IoT endpoints. IoT data is shared directly between two ‘peers’ and a server is only needed to enroll the devices in the P2P network for a direct connection. P2P IoT has much lower latency and higher privacy than database-driven IoT. However, current practices of addressing the wireless vulnerability by using the server to mediate a direct end-to-end encrypted connection between two IoT endpoints is too complex and inefficient. This technology offer is a lightweight and secure protocol for mutual authentication and key exchange directly between two endpoints in P2P IoT. The protocol exploits physically unclonable function (PUF) derived from manufacturing process variations of integrated circuits as device “biometrics”. The PUF circuit is lightweight and can be embedded in an endpoint device to generate unique, unpredictable, and unforgeable identity only upon query. PUF-based device identity is tamper-aware and hence more secure than hardcoded or memory-based identity. This PUF-based protocol is significantly more efficient and secure than cryptographic-key based protocols for P2P IoT applications. It enables any pair of IoT devices after enrollment to directly authenticate each other. Upon successful authentication, a secure and fresh shared session key is automatically established for encrypted communication, which directly overcomes the existing key distribution and management problem in P2P IoT.
Sweat Powered Flexible and Stretchable Printed Battery for Wearable Electronics
Wearable electronics is a US$116.2B dollar market that is only expected to grow exponentially with the increasing demand and use for wearable technologies in sectors such as healthcare and smart devices. One of the main limiting factors of this technology is the battery life which can be broken down to either having clunky charging capabilities or poor lasting time. Therefore, there is a need for either a better charging method or better performance for the battery.   This technology offer is a flexible and stretchable printed battery that provides seamless charging method via sweat. This battery is small (2x2cm) and can be charged with only 2mL of sweat to provide 20hrs worth of power. The battery can be used as either the primary or secondary source of power and can be printed on a variety of materials such as textiles. It contains no heavy metals or toxic chemicals which eliminates the problem of harmful electronic waste. The battery is also highly durable and can withstand strain from daily activities. Also, the slim nature of battery makes it less intrusive and allows for wider design options for wearable electronics.
Accelerated Materials Innovation Platform for Sustainability
Traditional materials research and development (R&D) is slow, challenging, expensive and therefore limited. The iterative nature limits companies in creating revolutionary products if they do not devote significant resources to R&D, often resulting in longer time to market of new products. This technology aims to resolve such issues by enabling fast and efficient materials development. Created with deep expertise in materials science, the technology utilises a proprietary combination of machine learning and high throughout experimentation as a method of acceleration for companies seeking to develop new materials. The platform allows users to develop new materials 10 times faster by performing rapid and smarter screening of data, design of experiments and materials creation. A key focus area the technology owner aims to create impact in is the development and application of materials that enable sustainability, particularly to resolve challenges in solving environmental waste, circular economy, energy storage, clean energy and carbon capture, utilization & storage. The company is seeking R&D projects and collaborations with interested parties to develop new materials.
Advanced Receiver for 5G/6G Network with High Linearity
The signal received by a radio frequency (RF) receiver is often accompanied by some blockers and interferers, which could be from in-band or out-of-band frequencies. Normally, they are with very large signal strength. As a result, they may easily jam and saturate the receiver. To alleviate their effect, one traditional solution is to utilize current-mode direct conversion. Another solution is to employ mixer-first receiver architecture. Although the current-mode and mixer-first architectures have certain tolerance for the blockers and interferers, they can still be saturated, especially with large blocker strength. This technology offer is an integrated circuit (IC) design of a novel true-current-mode receiver architecture, that can be used to alleviate the receiver saturation problem. The receiver start with a specially designed matching network. The matching networks have two options. Option-1 consists of R-L-C parallel paths, and option-2 is a passive 90° hybrid coupler. Both create “virtual ground” directly at the RF node, thus inducing voltage attenuation rather than voltage amplification at the RF node. This configuration improves the large-signal linearity greatly. Meanwhile, the noise of the matching resistor can be totally cancelled. Moreover, the local oscillator (LO) leakage to the RF port is greatly reduced. This technology offer is applicable to all mainstream communication systems, including, but not limited to, sub-6 GHz, 5G and 6G receivers. They are also applicable to surface acoustic wave (SAW)-less and full-duplex applications.
Novel Technique of Texturing Germanium for Anti-reflection in the Infrared Range
Anisotropic wet etching used to form inverted pyramid structures has been widely used in the complementary metal oxide semiconductor (CMOS) process of the silicon (Si) industry for the application of micro-electromechanical systems (MEMS) and optoelectronics. However, such CMOS-compatible anisotropic wet etching technique is still scarce for the germanium (Ge) industry. This technology offer is a technique to enable the formation of microscale Ge inverted pyramid and v-groove structures by wet etching, catalysed by CMOS-compatible metals. The technique has been proven feasible and the long-term durability in the etchant has also been verified. The dimensions of the Ge structures were totally determined by the patterned catalyst, which makes it easy for tuning desired sizes of Ge inverted structures. The Ge microscale textures show outstanding anti-reflective performance in the infrared (IR) range. The targeted users of this technique are Ge-based MEMS and optoelectronic device manufactures which requires CMOS-compatible fabrication flow to produce microscale structures with antireflective performance, but on Ge substrate. The technology owner is interested to out-license this process, or do research collaborations with foundries handling CMOS and photonics related processes.
Real Time, All-day, Stress Monitoring System Using Data Science
There are 30,000 occupational drivers in Singapore, out of which 13,500 are 45 years old and above. The risk of acquiring cardiovascular disease increases with age and is potentially exacerbated by low physical activity and high emotional stress levels, which are two typical characteristics of occupational drivers arising from their work environment. Low level of physical activity and high stress levels have been shown to have significant relationship with heart rate variability, one of the indicators of cardiovascular disease. This technology is developed to help drivers to monitor their stress level, provide them with instantaneous feedback and the necessary alerts for a timely intervention. This technology offer presents a cross-platform AI system that estimates the stress levels continuously in real time, and can be easily integrated with commercially available photoplethysmography (PPG) wearables, e.g., a PPG wristwatch. In addition, this technology can be adapted for the monitoring of workplace stress with the aim of improving overall mental well-being.
AI-Based Medical Imaging Assistant For Breast Cancer Screening
Breast cancer is one of the leading causes of mortality in the world and the most commonly occurring cancer in women. On average, one in 11 women will be affected by breast cancer in their lifetime, with more than 2.3 million women diagnosed in the year of 2020 alone. Early detection has the potential to save lives by significantly enhancing the survival rates as the chances of recovery drops significantly beyond stage 2 of the disease. Many countries have established screening programs for breast cancer as measures for early disease detection, as well as implemented relevant reimbursement schemes for supporting these programs. Among various screening methods, 2D mammography has been found to be the most accurate way of conducting such programs for large populations. However, there still lies challenges with existing mammography screening including inefficiencies in current clinical care workflow causing long waiting periods and inaccuracies in manual reading interpretation. The technology addresses these issues by augmenting the existing clinical workflow for radiologists diagnosing breast cancer by first, allowing for faster mammogram readings, and second eliminating the requirement for double-blind reading per screen for each diagnosis. The AI assistive technology tops in Asia for its proven AI algorithm for breast cancer detection at one of the highest AUC accuracy levels of 0.96, and is capable of reducing false positives for dense breast in Asian women.
Anti-Short Circuit Layer in Lithium-ion Batteries to Prevent Chemical Fires
Lithium-ion batteries is a US$40.5B dollar industry that has seen massive growth in recent years due to increasing demand for energy storage devices. However, one of the main issues with lithium-ion batteries is safety, with emphasis on chemical fires that is the result of short circuits caused by lithium dendrites built-up and growth. To date, this is still an existing problem with no viable solution offered yet. This technology offer is a proprietary anti-short circuit layer that can prevent chemical fires by stopping further lithium dendritic growth once these dendrites encounter the anti-short circuit layer. It has shown to be able to eliminate the risk of short-circuits completely for the entire life cycle of the battery. The application of the layer had no impact on the performance of the battery and is viable for almost all different compositions of lithium-ion batteries. Furthermore, addition of this layer only adds less than 5% (can be further reduced) to the total production cost with no changes to the manufacturing line.