Research

Integrated photonics faces significant barriers to innovation due to the high costs of design, prototyping, and testing. AI tools and automation can address these challenges, if the problems of hallucination and the lack of scientific reasoning can be overcome.  With collaborators, we are developing a platform rooted in physics and deductive reasoning that delivers precise, verifiable, and interpretable solutions. Unlike traditional large language models, the model avoids errors from associative reasoning and hallucinations, enabling accurate translation of complex engineering queries into mathematical formulations.

Our research focuses on applying this AI method to integrated photonics, accelerating development cycles and reducing costs to empower small teams of innovators. While we are focused on photonics, this platform has the potential to transform other deep-tech domains, including quantum computing and advanced electronics.

Axiomatic AI

As artificial intelligence continues to grow, the scalability, energy consumption, and environmental impact of machine learning systems have become critical challenges. Training large-scale neural networks faces computational bottlenecks, as the number of floating-point operations doubles every 10 months, while memory and network bandwidths increase only 20% per year. This results in significant costs and carbon emissions, underscoring the need for fundamentally new hardware that prioritizes energy efficiency and computational acceleration.

Our research focuses on leveraging integrated photonics to create next-generation computing systems. By connecting diverse processors (CPUs, GPUs, TPUs, and quantum processors) with a programmable photonic interconnect fabric and optoelectronic/electronic-optical interfaces, we aim to deliver unparalleled bandwidth, low latency, and energy efficiency. Central to this vision are innovations in photonic interconnects and opto-electronic interfaces, building on our prior work to address emerging paradigms in computing. 

We are a key pioneer in silicon nitride-on-silicon photonics and microring filters and modulators. These technologies have become core to the current commercial developments of siiicon photonics optical interconnects.

The field of silicon (Si) photonics aims to form photonic devices and circuits on Si substrates using the manufacturing infrastructure of CMOS electronics. The large Si substrates and wafer-scale microelectronic packaging lead to high production volumes that can reduce the cost of an optical chip. Si photonics is transforming data communications, which use infrared wavelengths near 1310nm or 1550nm.

We are working to create integrated photonic platforms on Si that operate in the visible wavelength range (between 450nm to 650nm).  The visible spectrum open numerous applications for integrated photonics, including micro-displays, quantum information, and sensing.  Silicon nitride waveguides with broadband transparency can be realized on Si substrates. However, many challenges abound, such as feature size limitations, material absorption, and active device (lasers and modulators) integration.  Many innovations are needed to realize highly functional visible light integrated photonic platforms.

We are bringing our expertise in foundry integrated photonics to create new technologies, such as implantable probes, recording arrays, and miniature microscopes, to study the brain. Much remains unknown about the complex and dense connectivity among the neurons in the brain. Developments in optogenetics, optical indicators, and multiphoton imaging are revolutionizing how light is used to study neural connectivity. Taking advantage of wafer-scale integration, our technologies will enable functionality that cannot be achieved with free-space or fiber optics, and will enable high density recording/stimulation in multiple modalities. These technologies are new tools for the neuroscience community and serve as new types interfaces to the brain.

Fiber optic communications can reduce the power consumption and latency in the communication within and between datacenters and are essential for broadband connectivity. Si photonics is a good candidate to serve high-volume markets due to the large Si wafer sizes, integration density, and mature manufacturing. 

To integrate lasers onto Si photonic platforms for the infrared spectrum, an approach is to bond InP-based optically amplifying materials onto the top surface of a Si photonic wafer.  In collaboration with SCINTIL Photonics, we are working on hybrid InP-on-Si integration where the InP is bonded from the backside of the wafer, onto the buried oxide (BOX) layer. Backside bonding enables hybrid laser integration onto multilayer Si photonic platforms and significantly improved wafer-scale uniformity in the oxide spacer thickness which should improve device yield.