3 or 4 decades ago, the invention of very large, inexpensive main memory architectures has been revolutionizing the market to develop large-scale computations performed resident in memory cached from dynamic RAM, attached to each processor chip. This has forever changed, how processors and chips accessing the program instructions and data from the DRAM instead of the spinning disks. A number of in-memory computing vendors have introduced machine learning and deep learning platforms that can process large-scale data in-memory such as GridGain Systems, MemSQL, QuanticMind, Gigaspaces, Hazelcast, Intel, Exasol, First derivates, Oracle, and IBM. Deep neural networks have advanced the field of AI and have achieved in performing the inference at low-energy for edge computing for autonomous vehicles, IoT-battery powered devices with DNN inference engines with in-memory computing architectures. In-memory memristive and resistive devices perform data transfers at scale on flash memory, and metal-oxide resistive RAM, and phase-change memory (PCM). The architectures of encoding the network weights as analog charge state in crossbar arrays, matrix-vector multiplications have powered the circuit laws and improved the mathematical precision calculations. Hyperdimensional in-memory computing is a novel emerging architecture that takes the attributes of neuronal circuits on non-Van Neumann paradigms that can perform a near-optimum trade-off in designing classification accuracy and design complexity with the mathematical properties of hyperdimensional spaces. Apache Ignite is a horizontally scalable and fault-tolerant distributed in-memory computing architecture that processes terabytes of big data at in-memory scale and speed, which is ACID-compliant with complete SQL support. GridGain is an in-memory computing platform for HPC environments that supports the data layers with SQL, NoSQL, zOS, and Hadoop.
