18-08-2020 | By Robin Mitchell
Intel has announced their change in operations to update chip architecture independently from silicon technology improvements. What problems have Intel faced, where is Moore’s law now, and what changes have Intel brought in?
To understand the difficulties faced by Intel, we first need to understand what Moore’s law is, and how it affects chip design. Microchips are small pieces of semiconductor that contain fully functional circuits which include resistors, capacitors, and transistors. The number of components that can fit onto a chip depends on the smallest feature that a fab house can produce (for example, a 250nm chip generally has the smallest feature being 250nm in size). Therefore, the smaller the feature size, the more components that can be put onto a single chip. The first chips (dating before the 1960s), would house up to 10 transistors, but modern designs can have as many as several billion. This trend of decreasing feature size and increasing component count was given a name, Moore’s Law, and states that the number of transistors on a chip double once every two years.
Moore’s law is an essential factor in modern electronics as it allows for more sophisticated devices to be produced. The shrinkage of transistors also reduces the power consumption of each device, which is critical for mobile applications. The ability to create more complex designs in the same space also helps with reducing the size of products, and it is this factor that has led to the widespread integration of IoT (Internet of Things), smart devices, and in general, modern life as we know it.
Since Moore’s Law greatly helps improve technology, keeping it going is an essential task for the semiconductor industry. However, Moore’s Law cannot work indefinitely, and transistors (constructed using atoms), can only be so small. The smallest feature sizes currently in mass production are 7nm, which is approximately 33 silicon atoms (assuming each silicon atoms van der val radius is 0.21mn), and these sizes already give designers headaches with the many effects that stop size reduction including quantum tunnelling and yield.
Intel is the world’s largest, most valuable semiconductor manufacturer, and as a result, is an industry leader in the field. Their success in their CPU line, which started with the Intel 4004, has set the gold standard to computing, and their x86 and x64 architecture are essentially common to all modern PCs and Laptops (mobile devices are often ARM-based). While the fundamentals of code architectures remain unchanged (as they need to retain backward compatibility), new features are always being added whether it’s a new hardware multiplier, faster DMA, or specialised page swapping features which can dramatically improve a computers speed of operation. Being a manufacturer, Intel likes to tie their advances in CPU technology with their advances in semiconductor technology, and thus new CPU features are often introduced when Intel reduces the sizes of transistors (known as a node). Therefore, the first x86 CPU may have been designed on a 250nm node, but the next version of x86 won’t be introduced until the 125nm node. This way, Intel CPU technology is tied into Moore’s Law.
While it may seem logical to tie the node size to new technology, this only works when a manufacturer can continue reducing the size of their node. This method also only works when the manufacturer can guarantee the next reduction on-time, something that Intel has historically not had a problem with. Of course, in the past, quantum effects were nowhere near as severe as they are in technologies under the 50nm mark, and moving from 14nm (Intel’s current node) to 7nm is a big challenge. While companies such as TMSC have managed to achieve 7nm, Intel has had multiple set-backs, and the result is Intel CPUs being two to three years behind in development.
AMD is a competitor to Intel who produces x86 and x64 compatible CPUs, while also producing APUs that integrate both the CPU and GPU into a single package. However, unlike Intel, AMD does not produce its own devices and instead outsources its manufacturing to other companies such as GlobalFoundries. As a result, AMD is primarily focused on improving its CPU technology, and integrating new hardware features that can help speed up processing times. Typically, Intel would be at the forefront of node technology thus generally being ahead of the competition, but with the delays in Intel’s latest tech, AMD has had a chance to overtake Intel with the releases of CPUs that are arguably more modern than Intel’s; something that has never happened. AMD has the advantage that if a manufacturer cannot reduce its feature size, then they can choose a different manufacturer who can. AMD also has the advantage that it does not need to be tied to node reduction, and can choose to adapt their newer designs to an older tech (such as sticking with 14nm).
As a result, Intel has recently announced that they will cut the tie between node technology and their CPU architecture. Raja Koduri, the former chief architect for AMD, was hired by Intel and identified the issues that Intel had been suffering. According to Raja Koduri, the inability to release newer architectures also prevents the company from reacting quickly to market demands, while also removing the ability to produce IP. The use of transistor-resilient designs will remove Intel’s dependence on Moore’s Law, and help to push out modern CPU architectures.