Chiplets: how to improve performance and guarantee calculation capacity
Chiplets are electronic components, or dies, whose properties enable the creation of smaller calculation systems, with more efficient communication and reduced production costs. Due to these features, they are now considered the future of the CPU sector.
In order to manage latest generation applications, IT teams must first find a way to maximise calculation capacity and minimise response times. Chiplets match these requirements, overcoming the physical constraints imposed by the size of transistors, which have now reached their miniaturisation limits.
Silica chiplets are the best option
Chiplets are constructed from a thin strip of semiconductor material, onto which a microprocessor circuit is installed. For many manufacturers, silica guarantees high quality standards and represents, without a doubt, the material which can offer the best performance.
According to results obtained so far, silica chiplets provide a reliable service because they are less susceptible to production defects, they have greater calculation capacity and relatively low production costs. They can also be included in complex architectures, ensuring high speed connections.
One example of this is 2D horizontal layout with 3D vertical connections among between logical chips. This architecture enables programmers to combine IP blocks and process technologies with memory elements and I/O. The main advantages offered by this type of interconnection are greater bandwidth and fewer latencies.
How chiplets are used in current CPUs
Nowadays, chiplets are often used in multicore CPUs, (consisting of several independent calculaton units) which work at the same time and allow different functionalities to operate together. The most common are dual-core and quad-core but, thanks to the continual improvement of construction technology, 6, 8, 10, 12 and 16 core processors are now available.
To further boost the calculaton capacity of microprocessors, multithreading can be used. It is a technique designed to allow the execution of several threads (virtual cores) on the same addressing space, which also share the same cache and the same translation lookaside buffer.
These two technologies can be combined, bringing significant advantages for system performance. The most modern CPUs implement both via the connection of virtual threads with a certain number of Physical cores (a processor with 4 cores in Hyper-Threading, for example can simulate the calculation capacity of 8 physical cores).
The constant technological evolution of chiplets has led to the need for a global standardisation system, so that maufacturers can produce their own chips. For this reason, in 2019, the working group Open Domain-Specific Architectures defined some sector specific standards and a practicable ecosystem.
Thanks to the new guidelines which make chiplets standard, producers can now use them on a single silica circuit, thereby obtaining a huge variety of CPU components which are also interchangeable.
Chiplets are therefore paving the way for the creation of modular processors based on ready to use system components. The main benefit of this approach is that the CPU can be specially made to support the calculation processes of a specific app.
Translated by Joanne Beckwith
