Disaggregation of system-on-chip (SOC) architecture
Embodiments described herein provide techniques to disaggregate an architecture of a system on a chip integrated circuit into multiple distinct chiplets that can be packaged onto a common chassis. In one embodiment, a graphics processing unit or parallel processor is composed from diverse silicon chiplets that are separately manufactured. A chiplet is an at least partially and distinctly packaged integrated circuit that includes distinct units of logic that can be assembled with other chiplets into a larger package. A diverse set of chiplets with different IP core logic can be assembled into a single device.
1. An apparatus comprising:
a package assembly including:
a first base chiplet comprising:
a first interconnect fabric, and
a first plurality of cache banks coupled to or integrated with the first interconnect fabric;
a first logic chiplet stacked on the first base chiplet, the first logic chiplet comprising:
a cluster of compute units to perform parallel execution of compute shader instructions or graphics shader instructions;
a first interconnect structure to couple the cluster of compute units to the first interconnect fabric;
a second base chiplet coupled to the first base chiplet by a second interconnect structure, the second base chiplet comprising:
a second interconnect fabric, and
a second plurality of cache banks coupled to or integrated with the second interconnect fabric;
a second logic chiplet stacked on the second base chiplet, the second logic chiplet comprising:
a plurality of application processor cores to execute instructions; and
a third interconnect structure to couple the second logic chiplet to the second interconnect fabric,
wherein the first logic chiplet is manufactured using a different process technology than that used to manufacture the first and second base chiplets.
2. The apparatus of claim 1 , wherein at least one of the first and second base chiplets further comprise:
a fourth interconnect structure to couple the at least one of the first and second base chiplets to a memory.
3. The apparatus of claim 2 , wherein the cluster of compute units and the plurality of application processor cores are to access the memory through the first and second base chiplets, respectively.
4. The apparatus of claim 3 , wherein the memory comprises a high-bandwidth memory (HBM).
5. The apparatus of claim 1 , further comprising:
a package substrate coupled to the first and second base chiplets, and
a package interconnect to electrically couple the package substrate to one or more devices.
6. The apparatus of claim 1 , wherein at least one of the first and second base chiplets further comprise:
a microcontroller to execute firmware to perform power management operations.
7. The apparatus of claim 6 , wherein the power management operations comprise controlling a voltage and frequency of at least one of the first and second base chiplets and the first and second logic chiplets.
8. The apparatus of claim 7 , wherein the voltage and frequency are to be independently controlled for the at least one of the first and second base chiplets and the first and second logic chiplets.
9. The apparatus of claim 7 , wherein each chiplet of the first and second base chiplets and the first and second logic chiplets comprises a separate power domain and clock domain.
10. A graphics processing system comprising:
a package assembly including:
a first base chiplet comprising:
a first interconnect fabric, and
a first plurality of cache banks coupled to or integrated with the first interconnect fabric;
a first logic chiplet stacked on the first base chiplet, the first logic chiplet comprising:
a cluster of compute units to perform parallel execution of compute shader instructions or graphics shader instructions;
a first interconnect structure to couple the cluster of compute units to the first interconnect fabric;
a second base chiplet coupled to the first base chiplet by a second interconnect structure, the second base chiplet comprising:
a second interconnect fabric, and
a second plurality of cache banks coupled to or integrated with the second interconnect fabric;
a second logic chiplet stacked on the second base chiplet, the second logic chiplet comprising:
a plurality of application processor cores to execute instructions; and
a third interconnect structure to couple the second logic chiplet to the second interconnect fabric,
wherein the first logic chiplet is manufactured using a different process technology than that used to manufacture the first and second base chiplets,
wherein at least one of the first and second base chiplets further comprise a fourth interconnect structure to couple the at least one of the first and second base chiplets to a memory, and
wherein the cluster of compute units and the plurality of application processor cores are to access the memory through the first and second base chiplets, respectively.
11. The graphics processing system of claim 10 wherein the memory comprises a high-bandwidth memory (HBM).
12. The graphics processing system of claim 10 further comprising:
a package substrate coupled to the first and second base chiplets, and
a package interconnect to electrically couple the package substrate to one or more devices.
13. The graphics processing system of claim 10 , wherein at least one of the first and second base chiplets further comprise:
a microcontroller to execute firmware to perform power management operations.
14. The graphics processing system of claim 13 , wherein the power management operations comprise controlling a voltage and frequency of at least one of the first and second base chiplets and the first and second logic chiplets.
15. The graphics processing system of claim 14 , wherein the voltage and frequency are to be independently controlled for the at least one of the first and second base chiplets and the first and second logic chiplets.
16. The graphics processing system of claim 14 , wherein each chiplet of the first and second base chiplets and the first and second logic chiplets comprises a separate power domain and clock domain.
17. A method of assembling an apparatus including a plurality of chiplets, the method comprising:
assembling a package assembly including a first base chiplet comprising a first interconnect fabric and a first plurality of cache banks coupled to or integrated with the first interconnect fabric;
stacking a first logic chiplet on the first base chiplet, the first logic chiplet comprising a cluster of compute units to perform parallel execution of compute shader instructions or graphics shader instructions;
coupling the cluster of compute units to the first interconnect fabric via a first interconnect structure;
coupling the first base chiplet to a second base chiplet by a second interconnect structure, the second base chiplet comprising a second interconnect fabric and a second plurality of cache banks coupled to or integrated with the second interconnect fabric;
stacking a second logic chiplet on the second base chiplet, the second logic chiplet comprising a plurality of application processor cores to execute instructions; and
coupling the second logic chiplet to the second interconnect fabric via a third interconnect structure, wherein the first logic chiplet is manufactured using a different process technology than that used to manufacture the first base chiplet and the second base chiplet.
18. The method of claim 17 , further comprising coupling the at least one of the first and second base chiplets to a memory via a fourth interconnect structure and enabling the cluster of compute units and the plurality of application processor cores to access the memory through the first and second base chiplets, respectively, wherein the memory comprises a high-bandwidth memory (HBM).
19. The method of claim 18 , further comprising coupling a package substrate to the first and second base chiplets and electrically coupling the package substrate to one or more devices via a package interconnect.
20. The method of claim 17 , further comprising configuring a microcontroller of at least one of the first and second base chiplets to execute firmware to perform power management operations, the power management operations comprising controlling a voltage and frequency of at least one of the first and second base chiplets and the first and second logic chiplets.