GEMINI-II APU TO TARGET SMALL MODEL ALGORITHMS THAT FIT ENTIRELY INTO THE CHIP'S MEMORY FOR EDGE APPLICATIONS SUNNYVALE, Calif., May 09, 2024 (GLOBE NEWSWIRE) -- GSI Technology, Inc. (NASDAQ:GSIT), the inventor of the Associative Processing Unit (APU), a paradigm shift in AI and HPC processing providing true compute-in-memory technology, today reiterated the key highlights from its recent earnings call for the fourth quarter of fiscal year 2024 held on Thursday, May 2, 2024. Key Business Updates from Earnings Call Launched two high-capacity, low-power 1U and 2U servers integrated with the High-performance Gemini-I APU and designed specifically for SAR (for potential applications in tracking troop and equipment movement, forest fires) and Fast Vector Search (for possible applications in e-commerce, facial recognition, and molecular search) applications, enabling mobile applications and enterprise-level processing at the edge. In addition, Gemini-I will serve as a mechanism to demonstrate a solution for small models that incorporate BitNet and Binary Neural Networks (BNN) for Gemini-II market adoption. Announced a Radiation Hardened shipment for a European Space Agency (ESA) mission. This is the first Rad Hard shipment for a non-U.S.-based mission, giving GSI exposure within the ESA community. Disclosed the sale and lease-back of its headquarters for $11.9 million, expected to close in June. The proceeds will boost the Company's cash position with additional funding to support the finalization of Gemini-II and other R&D projects. Announced initiating a broad strategic review to maximize stockholder value and retained Needham & Company, LLC as the strategic and financial advisor. Key Highlights from Gemini-II Update on Earnings Call Gemini-II is currently undergoing rigorous testing and debugging, including the integration of the chip onto a board, which has enabled comprehensive performance assessments and produced results that have exceeded initial expectations. GSI's software team is actively writing libraries to develop new applications on the edge or near the edge for Gemini-II. Its substantial processing capabilities can empower the local execution of computationally intensive tasks to increase edge application capabilities like advanced driver assistance systems for automobiles and HPC in delivery drones, autonomous robots, unmanned aerial vehicles, and satellites. Gemini-II's memory can hold a small database, a potential door opener for enhanced performance in several applications. One example would be an off-the-shelf facial recognition solution, potentially in hardware with on-prem software or SaaS. "Engaging Needham & Company underscores our proactive approach to fortifying our market position and financial health and maximizing stockholder value," said Lee-Lean Shu, CEO of GSI Technology. "We are making excellent progress with our second generation of the APU, Gemini-II, and are extremely encouraged by the chip's results in testing, which has verified that instructions can be successfully executed through the embedded processor and that the data path is also working. Once we receive the second spin of Gemini-II this fall, we intend to initiate benchmarking that will allow us to begin preliminary customer sampling and engage with target customers before the calendar year-end. "The second generation of our APU brings significant performance enhancements, with more processing power and memory density, and suitable for both low power data center expansion and enabling data center functions at the edge," continued Mr. Shu. "These capabilities empower local execution of computationally intensive tasks, increasing edge application capabilities. Additionally, Gemini-II's memory can hold an entire small database, a potential door-opener for enhanced performance in image recognition applications. These small model sizes use algorithms with low precision model weight size to lower memory storage requirements and simplify computations, which the APU architecture is better suited for than traditional GPU architectures. We plan to demonstrate that the entire ...